Hugging Face's logo

bong9513
/
Visibility_Nowcasting

Model card Files
xet
 Community
3
bong9513 commited on
Commit
d3488a7
·
verified ·
1 Parent(s): 4ac637a

Delete Analysis_code

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. Analysis_code/0.air_data_merge.ipynb +0 -1029
  2. Analysis_code/1.data_merge.ipynb +0 -0
  3. Analysis_code/2.eda_preproccesing.ipynb +0 -3
  4. Analysis_code/3.oversampling.ipynb +0 -0
  5. Analysis_code/__pycache__/code.cpython-312.pyc +0 -0
  6. Analysis_code/__pycache__/deepgbm.cpython-312.pyc +0 -0
  7. Analysis_code/__pycache__/deepgbm.cpython-38.pyc +0 -0
  8. Analysis_code/__pycache__/deepgbm.cpython-39.pyc +0 -0
  9. Analysis_code/__pycache__/ft_transformer.cpython-312.pyc +0 -0
  10. Analysis_code/__pycache__/ft_transformer.cpython-38.pyc +0 -0
  11. Analysis_code/__pycache__/ft_transformer.cpython-39.pyc +0 -0
  12. Analysis_code/__pycache__/resnet_like.cpython-312.pyc +0 -0
  13. Analysis_code/__pycache__/resnet_like.cpython-38.pyc +0 -0
  14. Analysis_code/__pycache__/resnet_like.cpython-39.pyc +0 -0
  15. Analysis_code/best_deepgbm_model.pth +0 -3
  16. Analysis_code/best_model_f1.pth +0 -3
  17. Analysis_code/deepgbm.py +0 -47
  18. Analysis_code/deeplearning_model_binary.ipynb +0 -0
  19. Analysis_code/deeplearning_model_multi.ipynb +0 -0
  20. Analysis_code/final_test/final.ipynb +0 -1143
  21. Analysis_code/find_reason/ busan_trend.ipynb +0 -0
  22. Analysis_code/find_reason/ daegu_trend.ipynb +0 -0
  23. Analysis_code/find_reason/ gwangju_trend.ipynb +0 -0
  24. Analysis_code/find_reason/ incheon_trend.ipynb +0 -0
  25. Analysis_code/find_reason/ seoul_trend.ipynb +0 -0
  26. Analysis_code/find_reason/daejeon_trend.ipynb +0 -0
  27. Analysis_code/find_reason/make_trend_plot.ipynb +0 -0
  28. Analysis_code/find_reason/wasserstein_distance.ipynb +0 -541
  29. Analysis_code/ft_transformer.py +0 -56
  30. Analysis_code/make_oversample_data/gan_sample_10000_1.py +0 -181
  31. Analysis_code/make_oversample_data/gan_sample_10000_2.py +0 -182
  32. Analysis_code/make_oversample_data/gan_sample_10000_3.py +0 -182
  33. Analysis_code/make_oversample_data/gan_sample_20000_1.py +0 -183
  34. Analysis_code/make_oversample_data/gan_sample_20000_2.py +0 -183
  35. Analysis_code/make_oversample_data/gan_sample_20000_3.py +0 -183
  36. Analysis_code/make_oversample_data/gan_sample_7000_1.py +0 -180
  37. Analysis_code/make_oversample_data/gan_sample_7000_2.py +0 -182
  38. Analysis_code/make_oversample_data/gan_sample_7000_3.py +0 -182
  39. Analysis_code/make_oversample_data/oversampling_code.py +0 -355
  40. Analysis_code/make_oversample_data/smote_sample_1.py +0 -53
  41. Analysis_code/make_oversample_data/smote_sample_2.py +0 -53
  42. Analysis_code/make_oversample_data/smote_sample_3.py +0 -53
  43. Analysis_code/make_train_test.ipynb +0 -1099
  44. Analysis_code/model_result/best_sample/ensemble_best_sample.csv +0 -157
  45. Analysis_code/model_result/deepgbm_sampled_data_test.csv +0 -31
  46. Analysis_code/model_result/ft_transformer_sampled_data_test.csv +0 -31
  47. Analysis_code/model_result/lightgbm_sampled_data_test.csv +0 -31
  48. Analysis_code/model_result/resnet_like_sampled_data_test.csv +0 -31
  49. Analysis_code/model_result/xgboost_sampled_data_test.csv +0 -31
  50. Analysis_code/model_visualize.ipynb +0 -0
Analysis_code/0.air_data_merge.ipynb DELETED
@@ -1,1029 +0,0 @@
1
- {
2
- "cells": [
3
- {
4
- "cell_type": "code",
5
- "execution_count": 1,
6
- "metadata": {},
7
- "outputs": [
8
- {
9
- "name": "stderr",
10
- "output_type": "stream",
11
- "text": [
12
- "/opt/conda/lib/python3.8/site-packages/tqdm/auto.py:22: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
13
- " from .autonotebook import tqdm as notebook_tqdm\n"
14
- ]
15
- }
16
- ],
17
- "source": [
18
- "import os\n",
19
- "import numpy as np\n",
20
- "import pandas as pd\n",
21
- "import natsort\n",
22
- "from datetime import datetime\n",
23
- "from tqdm.auto import tqdm"
24
- ]
25
- },
26
- {
27
- "cell_type": "code",
28
- "execution_count": 2,
29
- "metadata": {},
30
- "outputs": [],
31
- "source": [
32
- "def get_data(year):\n",
33
- " files = natsort.natsorted(os.listdir(f'../data/대기질/{year}/'))\n",
34
- " data = []\n",
35
- " for file in tqdm(files, desc=f\"Reading files...({len(files)})\"):\n",
36
- " data.append(pd.read_excel(f'../data/대기질/{year}/{file}', usecols=[\"지역\", '망', \"측정소코드\", \"측정소명\", \"측정일시\", \"O3\", \"NO2\", \"PM10\", \"PM25\", \"주소\"]))\n",
37
- "\n",
38
- " return pd.concat(data)"
39
- ]
40
- },
41
- {
42
- "cell_type": "code",
43
- "execution_count": 3,
44
- "metadata": {},
45
- "outputs": [],
46
- "source": [
47
- "# 합친 데이터에 날짜 정보를 추가한다.\n",
48
- "def add_date(df):\n",
49
- "\n",
50
- " df[\"측정일시\"] = df[\"측정일시\"].astype(str).str[:10]\n",
51
- " df[\"측정일시\"] = pd.to_datetime(df[\"측정일시\"], format='%Y%m%d%H', errors=\"coerce\")\n",
52
- "\n",
53
- " df[\"year\"] = df[\"측정일시\"].dt.year\n",
54
- " df[\"month\"] = df[\"측정일시\"].dt.month\n",
55
- " df[\"day\"] = df[\"측정일시\"].dt.day\n",
56
- " df[\"hour\"] = df[\"측정일시\"].dt.hour\n",
57
- "\n",
58
- " return df"
59
- ]
60
- },
61
- {
62
- "cell_type": "code",
63
- "execution_count": 4,
64
- "metadata": {},
65
- "outputs": [
66
- {
67
- "name": "stderr",
68
- "output_type": "stream",
69
- "text": [
70
- " 0%| | 0/6 [00:00<?, ?it/s]\n",
71
- "Reading files...(13): 0%| | 0/13 [00:00<?, ?it/s]\u001b[A\n",
72
- "Reading files...(13): 8%|▊ | 1/13 [00:34<06:57, 34.80s/it]\u001b[A\n",
73
- "Reading files...(13): 15%|█▌ | 2/13 [01:12<06:41, 36.47s/it]\u001b[A\n",
74
- "Reading files...(13): 23%|██▎ | 3/13 [01:47<05:58, 35.89s/it]\u001b[A\n",
75
- "Reading files...(13): 31%|███ | 4/13 [02:23<05:23, 35.96s/it]\u001b[A\n",
76
- "Reading files...(13): 38%|███▊ | 5/13 [02:59<04:47, 35.92s/it]\u001b[A\n",
77
- "Reading files...(13): 46%|████▌ | 6/13 [03:35<04:12, 36.09s/it]\u001b[A\n",
78
- "Reading files...(13): 62%|██████▏ | 8/13 [04:12<02:16, 27.35s/it]\u001b[A\n",
79
- "Reading files...(13): 69%|██████▉ | 9/13 [04:46<01:56, 29.05s/it]\u001b[A\n",
80
- "Reading files...(13): 77%|███████▋ | 10/13 [05:21<01:31, 30.55s/it]\u001b[A\n",
81
- "Reading files...(13): 85%|████████▍ | 11/13 [05:58<01:04, 32.46s/it]\u001b[A\n",
82
- "Reading files...(13): 92%|█████████▏| 12/13 [06:37<00:34, 34.28s/it]\u001b[A\n",
83
- "Reading files...(13): 100%|██████████| 13/13 [07:08<00:00, 32.93s/it]\u001b[A\n",
84
- " 17%|█▋ | 1/6 [07:18<36:30, 438.18s/it]\n",
85
- "Reading files...(13): 0%| | 0/13 [00:00<?, ?it/s]\u001b[A\n",
86
- "Reading files...(13): 8%|▊ | 1/13 [00:43<08:41, 43.43s/it]\u001b[A\n",
87
- "Reading files...(13): 15%|█▌ | 2/13 [01:26<07:56, 43.29s/it]\u001b[A\n",
88
- "Reading files...(13): 23%|██▎ | 3/13 [02:07<07:02, 42.22s/it]\u001b[A\n",
89
- "Reading files...(13): 31%|███ | 4/13 [02:50<06:23, 42.66s/it]\u001b[A\n",
90
- "Reading files...(13): 38%|███▊ | 5/13 [03:28<05:27, 40.90s/it]\u001b[A\n",
91
- "Reading files...(13): 46%|████▌ | 6/13 [04:15<04:59, 42.79s/it]\u001b[A\n",
92
- "Reading files...(13): 54%|█████▍ | 7/13 [04:58<04:18, 43.14s/it]\u001b[A\n",
93
- "Reading files...(13): 62%|██████▏ | 8/13 [05:43<03:37, 43.47s/it]\u001b[A\n",
94
- "Reading files...(13): 69%|██████▉ | 9/13 [06:28<02:55, 43.96s/it]\u001b[A\n",
95
- "Reading files...(13): 77%|███████▋ | 10/13 [07:12<02:12, 44.01s/it]\u001b[A\n",
96
- "Reading files...(13): 85%|████████▍ | 11/13 [07:52<01:25, 42.90s/it]\u001b[A\n",
97
- "Reading files...(13): 100%|██████████| 13/13 [08:34<00:00, 39.61s/it]\u001b[A\n",
98
- " 33%|███▎ | 2/6 [16:05<32:42, 490.55s/it]\n",
99
- "Reading files...(13): 0%| | 0/13 [00:00<?, ?it/s]\u001b[A\n",
100
- "Reading files...(13): 8%|▊ | 1/13 [00:49<09:56, 49.74s/it]\u001b[A\n",
101
- "Reading files...(13): 15%|█▌ | 2/13 [01:43<09:31, 51.98s/it]\u001b[A\n",
102
- "Reading files...(13): 23%|██▎ | 3/13 [02:33<08:29, 50.96s/it]\u001b[A\n",
103
- "Reading files...(13): 31%|███ | 4/13 [03:23<07:38, 50.95s/it]\u001b[A\n",
104
- "Reading files...(13): 38%|███▊ | 5/13 [04:13<06:43, 50.46s/it]\u001b[A\n",
105
- "Reading files...(13): 46%|████▌ | 6/13 [04:58<05:40, 48.71s/it]\u001b[A\n",
106
- "Reading files...(13): 54%|█████▍ | 7/13 [05:50<04:57, 49.66s/it]\u001b[A\n",
107
- "Reading files...(13): 62%|██████▏ | 8/13 [06:45<04:16, 51.29s/it]\u001b[A\n",
108
- "Reading files...(13): 77%|███████▋ | 10/13 [07:38<01:58, 39.46s/it]\u001b[A\n",
109
- "Reading files...(13): 85%|████████▍ | 11/13 [08:30<01:25, 42.79s/it]\u001b[A\n",
110
- "Reading files...(13): 92%|█████████▏| 12/13 [09:26<00:46, 46.32s/it]\u001b[A\n",
111
- "Reading files...(13): 100%|██████████| 13/13 [10:13<00:00, 47.19s/it]\u001b[A\n",
112
- " 50%|█████ | 3/6 [26:32<27:38, 552.96s/it]\n",
113
- "Reading files...(13): 0%| | 0/13 [00:00<?, ?it/s]\u001b[A\n",
114
- "Reading files...(13): 8%|▊ | 1/13 [00:59<11:48, 59.01s/it]\u001b[A\n",
115
- "Reading files...(13): 15%|█▌ | 2/13 [01:56<10:40, 58.19s/it]\u001b[A\n",
116
- "Reading files...(13): 23%|██▎ | 3/13 [02:53<09:37, 57.77s/it]\u001b[A\n",
117
- "Reading files...(13): 31%|███ | 4/13 [03:52<08:41, 58.00s/it]\u001b[A\n",
118
- "Reading files...(13): 38%|███▊ | 5/13 [04:44<07:26, 55.77s/it]\u001b[A\n",
119
- "Reading files...(13): 46%|████▌ | 6/13 [05:40<06:32, 56.05s/it]\u001b[A\n",
120
- "Reading files...(13): 54%|█████▍ | 7/13 [06:36<05:36, 56.06s/it]\u001b[A\n",
121
- "Reading files...(13): 62%|██████▏ | 8/13 [07:33<04:42, 56.42s/it]\u001b[A\n",
122
- "Reading files...(13): 69%|██████▉ | 9/13 [08:34<03:51, 57.76s/it]\u001b[A\n",
123
- "Reading files...(13): 77%|███████▋ | 10/13 [09:35<02:56, 58.75s/it]\u001b[A\n",
124
- "Reading files...(13): 92%|█████████▏| 12/13 [10:33<00:44, 44.84s/it]\u001b[A\n",
125
- "Reading files...(13): 100%|██████████| 13/13 [11:32<00:00, 53.29s/it]\u001b[A\n",
126
- " 67%|██████▋ | 4/6 [38:20<20:28, 614.26s/it]\n",
127
- "Reading files...(13): 0%| | 0/13 [00:00<?, ?it/s]\u001b[A\n",
128
- "Reading files...(13): 8%|▊ | 1/13 [00:59<11:57, 59.79s/it]\u001b[A\n",
129
- "Reading files...(13): 15%|█▌ | 2/13 [02:01<11:07, 60.67s/it]\u001b[A\n",
130
- "Reading files...(13): 23%|██▎ | 3/13 [03:02<10:10, 61.02s/it]\u001b[A\n",
131
- "Reading files...(13): 31%|███ | 4/13 [03:57<08:48, 58.74s/it]\u001b[A\n",
132
- "Reading files...(13): 38%|███▊ | 5/13 [04:57<07:53, 59.18s/it]\u001b[A\n",
133
- "Reading files...(13): 46%|████▌ | 6/13 [06:00<07:03, 60.45s/it]\u001b[A\n",
134
- "Reading files...(13): 54%|█████▍ | 7/13 [07:00<06:02, 60.38s/it]\u001b[A\n",
135
- "Reading files...(13): 62%|██████▏ | 8/13 [08:02<05:04, 60.85s/it]\u001b[A\n",
136
- "Reading files...(13): 69%|██████▉ | 9/13 [09:04<04:04, 61.03s/it]\u001b[A\n",
137
- "Reading files...(13): 77%|███████▋ | 10/13 [10:04<03:02, 60.67s/it]\u001b[A\n",
138
- "Reading files...(13): 92%|█████████▏| 12/13 [11:06<00:46, 46.76s/it]\u001b[A\n",
139
- "Reading files...(13): 100%|██████████| 13/13 [12:09<00:00, 56.08s/it]\u001b[A\n",
140
- " 83%|████████▎ | 5/6 [50:46<11:01, 661.78s/it]\n",
141
- "Reading files...(13): 0%| | 0/13 [00:00<?, ?it/s]\u001b[A\n",
142
- "Reading files...(13): 8%|▊ | 1/13 [01:03<12:46, 63.88s/it]\u001b[A\n",
143
- "Reading files...(13): 15%|█▌ | 2/13 [02:08<11:50, 64.56s/it]\u001b[A\n",
144
- "Reading files...(13): 23%|██▎ | 3/13 [03:10<10:32, 63.22s/it]\u001b[A\n",
145
- "Reading files...(13): 31%|███ | 4/13 [04:07<09:05, 60.63s/it]\u001b[A\n",
146
- "Reading files...(13): 38%|███▊ | 5/13 [05:09<08:11, 61.41s/it]\u001b[A\n",
147
- "Reading files...(13): 46%|████▌ | 6/13 [06:12<07:13, 61.92s/it]\u001b[A\n",
148
- "Reading files...(13): 54%|█████▍ | 7/13 [07:13<06:09, 61.50s/it]\u001b[A\n",
149
- "Reading files...(13): 62%|██████▏ | 8/13 [08:15<05:08, 61.64s/it]\u001b[A\n",
150
- "Reading files...(13): 69%|██████▉ | 9/13 [09:17<04:07, 61.81s/it]\u001b[A\n",
151
- "Reading files...(13): 77%|███████▋ | 10/13 [10:19<03:05, 61.96s/it]\u001b[A\n",
152
- "Reading files...(13): 92%|█████████▏| 12/13 [11:23<00:47, 47.75s/it]\u001b[A\n",
153
- "Reading files...(13): 100%|██████████| 13/13 [12:27<00:00, 57.50s/it]\u001b[A\n",
154
- "100%|██████████| 6/6 [1:03:31<00:00, 635.28s/it]\n"
155
- ]
156
- }
157
- ],
158
- "source": [
159
- "import os\n",
160
- "import pandas as pd\n",
161
- "from tqdm.auto import tqdm\n",
162
- "\n",
163
- "# 대기질 데이터를 불러와서 하나의 파일로 합친다.\n",
164
- "def get_data(year):\n",
165
- " directory = f'../data/대기질/{year}/'\n",
166
- " files = os.listdir(directory)\n",
167
- " data = []\n",
168
- " \n",
169
- " # 파일 목록에서 디렉토리를 제외하고 오직 Excel 파일만 처리\n",
170
- " for file in tqdm(files, desc=f\"Reading files...({len(files)})\"):\n",
171
- " file_path = os.path.join(directory, file)\n",
172
- " if os.path.isfile(file_path) and file_path.endswith(('.xls', '.xlsx')): # Excel 파일 확장자만 허용\n",
173
- " data.append(pd.read_excel(file_path, usecols=[\"지역\", '망', \"측정소코드\", \"측정소명\", \"측정일시\", \"O3\", \"NO2\", \"PM10\", \"PM25\", \"주소\"]))\n",
174
- " \n",
175
- " return pd.concat(data)\n",
176
- "\n",
177
- "years = [2018, 2019, 2020,2021,2022,2023] # 2018년부터 2023년까지의 데이터를 합친다.\n",
178
- "for year in tqdm(years):\n",
179
- " data = get_data(year)\n",
180
- " data = add_date(data)\n",
181
- " data.reset_index(drop=True, inplace=True)\n",
182
- " data.to_feather(f\"../data/대기질/{year}.feather\")\n"
183
- ]
184
- },
185
- {
186
- "cell_type": "code",
187
- "execution_count": 6,
188
- "metadata": {},
189
- "outputs": [
190
- {
191
- "data": {
192
- "text/html": [
193
- "<div>\n",
194
- "<style scoped>\n",
195
- " .dataframe tbody tr th:only-of-type {\n",
196
- " vertical-align: middle;\n",
197
- " }\n",
198
- "\n",
199
- " .dataframe tbody tr th {\n",
200
- " vertical-align: top;\n",
201
- " }\n",
202
- "\n",
203
- " .dataframe thead th {\n",
204
- " text-align: right;\n",
205
- " }\n",
206
- "</style>\n",
207
- "<table border=\"1\" class=\"dataframe\">\n",
208
- " <thead>\n",
209
- " <tr style=\"text-align: right;\">\n",
210
- " <th></th>\n",
211
- " <th>지역</th>\n",
212
- " <th>망</th>\n",
213
- " <th>측정소코드</th>\n",
214
- " <th>측정소명</th>\n",
215
- " <th>측정일시</th>\n",
216
- " <th>O3</th>\n",
217
- " <th>NO2</th>\n",
218
- " <th>PM10</th>\n",
219
- " <th>PM25</th>\n",
220
- " <th>주소</th>\n",
221
- " <th>year</th>\n",
222
- " <th>month</th>\n",
223
- " <th>day</th>\n",
224
- " <th>hour</th>\n",
225
- " </tr>\n",
226
- " </thead>\n",
227
- " <tbody>\n",
228
- " <tr>\n",
229
- " <th>0</th>\n",
230
- " <td>서울 중구</td>\n",
231
- " <td>도시대기</td>\n",
232
- " <td>111121</td>\n",
233
- " <td>중구</td>\n",
234
- " <td>2023-07-01 01:00:00</td>\n",
235
- " <td>0.0249</td>\n",
236
- " <td>0.0188</td>\n",
237
- " <td>21.0</td>\n",
238
- " <td>19.0</td>\n",
239
- " <td>서울 중구 덕수궁길 15</td>\n",
240
- " <td>2023.0</td>\n",
241
- " <td>7.0</td>\n",
242
- " <td>1.0</td>\n",
243
- " <td>1.0</td>\n",
244
- " </tr>\n",
245
- " <tr>\n",
246
- " <th>1</th>\n",
247
- " <td>서울 중구</td>\n",
248
- " <td>도시대기</td>\n",
249
- " <td>111121</td>\n",
250
- " <td>중구</td>\n",
251
- " <td>2023-07-01 02:00:00</td>\n",
252
- " <td>0.0263</td>\n",
253
- " <td>0.0163</td>\n",
254
- " <td>18.0</td>\n",
255
- " <td>15.0</td>\n",
256
- " <td>서울 중구 덕수궁길 15</td>\n",
257
- " <td>2023.0</td>\n",
258
- " <td>7.0</td>\n",
259
- " <td>1.0</td>\n",
260
- " <td>2.0</td>\n",
261
- " </tr>\n",
262
- " <tr>\n",
263
- " <th>2</th>\n",
264
- " <td>서울 중구</td>\n",
265
- " <td>도시대기</td>\n",
266
- " <td>111121</td>\n",
267
- " <td>중구</td>\n",
268
- " <td>2023-07-01 03:00:00</td>\n",
269
- " <td>0.0218</td>\n",
270
- " <td>0.0192</td>\n",
271
- " <td>24.0</td>\n",
272
- " <td>21.0</td>\n",
273
- " <td>서울 중구 덕수궁길 15</td>\n",
274
- " <td>2023.0</td>\n",
275
- " <td>7.0</td>\n",
276
- " <td>1.0</td>\n",
277
- " <td>3.0</td>\n",
278
- " </tr>\n",
279
- " <tr>\n",
280
- " <th>3</th>\n",
281
- " <td>서울 중구</td>\n",
282
- " <td>도시대기</td>\n",
283
- " <td>111121</td>\n",
284
- " <td>중구</td>\n",
285
- " <td>2023-07-01 04:00:00</td>\n",
286
- " <td>0.0131</td>\n",
287
- " <td>0.0214</td>\n",
288
- " <td>25.0</td>\n",
289
- " <td>19.0</td>\n",
290
- " <td>서울 중구 덕수궁길 15</td>\n",
291
- " <td>2023.0</td>\n",
292
- " <td>7.0</td>\n",
293
- " <td>1.0</td>\n",
294
- " <td>4.0</td>\n",
295
- " </tr>\n",
296
- " <tr>\n",
297
- " <th>4</th>\n",
298
- " <td>서울 중구</td>\n",
299
- " <td>도시대기</td>\n",
300
- " <td>111121</td>\n",
301
- " <td>중구</td>\n",
302
- " <td>2023-07-01 05:00:00</td>\n",
303
- " <td>0.0131</td>\n",
304
- " <td>0.0160</td>\n",
305
- " <td>25.0</td>\n",
306
- " <td>21.0</td>\n",
307
- " <td>서울 중구 덕수궁길 15</td>\n",
308
- " <td>2023.0</td>\n",
309
- " <td>7.0</td>\n",
310
- " <td>1.0</td>\n",
311
- " <td>5.0</td>\n",
312
- " </tr>\n",
313
- " <tr>\n",
314
- " <th>5</th>\n",
315
- " <td>서울 중구</td>\n",
316
- " <td>도시대기</td>\n",
317
- " <td>111121</td>\n",
318
- " <td>중구</td>\n",
319
- " <td>2023-07-01 06:00:00</td>\n",
320
- " <td>0.0115</td>\n",
321
- " <td>0.0196</td>\n",
322
- " <td>23.0</td>\n",
323
- " <td>18.0</td>\n",
324
- " <td>서울 중구 덕수궁길 15</td>\n",
325
- " <td>2023.0</td>\n",
326
- " <td>7.0</td>\n",
327
- " <td>1.0</td>\n",
328
- " <td>6.0</td>\n",
329
- " </tr>\n",
330
- " <tr>\n",
331
- " <th>6</th>\n",
332
- " <td>서울 중구</td>\n",
333
- " <td>도시대기</td>\n",
334
- " <td>111121</td>\n",
335
- " <td>중구</td>\n",
336
- " <td>2023-07-01 07:00:00</td>\n",
337
- " <td>0.0094</td>\n",
338
- " <td>0.0230</td>\n",
339
- " <td>26.0</td>\n",
340
- " <td>21.0</td>\n",
341
- " <td>서울 중구 덕수궁길 15</td>\n",
342
- " <td>2023.0</td>\n",
343
- " <td>7.0</td>\n",
344
- " <td>1.0</td>\n",
345
- " <td>7.0</td>\n",
346
- " </tr>\n",
347
- " <tr>\n",
348
- " <th>7</th>\n",
349
- " <td>서울 중구</td>\n",
350
- " <td>도시대기</td>\n",
351
- " <td>111121</td>\n",
352
- " <td>중구</td>\n",
353
- " <td>2023-07-01 08:00:00</td>\n",
354
- " <td>0.0222</td>\n",
355
- " <td>0.0175</td>\n",
356
- " <td>26.0</td>\n",
357
- " <td>20.0</td>\n",
358
- " <td>서울 중구 덕수궁길 15</td>\n",
359
- " <td>2023.0</td>\n",
360
- " <td>7.0</td>\n",
361
- " <td>1.0</td>\n",
362
- " <td>8.0</td>\n",
363
- " </tr>\n",
364
- " <tr>\n",
365
- " <th>8</th>\n",
366
- " <td>서울 중구</td>\n",
367
- " <td>도시대기</td>\n",
368
- " <td>111121</td>\n",
369
- " <td>중구</td>\n",
370
- " <td>2023-07-01 09:00:00</td>\n",
371
- " <td>0.0396</td>\n",
372
- " <td>0.0153</td>\n",
373
- " <td>27.0</td>\n",
374
- " <td>20.0</td>\n",
375
- " <td>서울 중구 덕수궁길 15</td>\n",
376
- " <td>2023.0</td>\n",
377
- " <td>7.0</td>\n",
378
- " <td>1.0</td>\n",
379
- " <td>9.0</td>\n",
380
- " </tr>\n",
381
- " <tr>\n",
382
- " <th>9</th>\n",
383
- " <td>서울 중구</td>\n",
384
- " <td>도시대기</td>\n",
385
- " <td>111121</td>\n",
386
- " <td>중구</td>\n",
387
- " <td>2023-07-01 10:00:00</td>\n",
388
- " <td>0.0530</td>\n",
389
- " <td>0.0105</td>\n",
390
- " <td>19.0</td>\n",
391
- " <td>16.0</td>\n",
392
- " <td>서울 중구 덕수궁길 15</td>\n",
393
- " <td>2023.0</td>\n",
394
- " <td>7.0</td>\n",
395
- " <td>1.0</td>\n",
396
- " <td>10.0</td>\n",
397
- " </tr>\n",
398
- " <tr>\n",
399
- " <th>10</th>\n",
400
- " <td>서울 중구</td>\n",
401
- " <td>도시대기</td>\n",
402
- " <td>111121</td>\n",
403
- " <td>중구</td>\n",
404
- " <td>2023-07-01 11:00:00</td>\n",
405
- " <td>0.0607</td>\n",
406
- " <td>0.0090</td>\n",
407
- " <td>20.0</td>\n",
408
- " <td>20.0</td>\n",
409
- " <td>서울 중구 덕수궁길 15</td>\n",
410
- " <td>2023.0</td>\n",
411
- " <td>7.0</td>\n",
412
- " <td>1.0</td>\n",
413
- " <td>11.0</td>\n",
414
- " </tr>\n",
415
- " <tr>\n",
416
- " <th>11</th>\n",
417
- " <td>서울 중구</td>\n",
418
- " <td>도시대기</td>\n",
419
- " <td>111121</td>\n",
420
- " <td>중구</td>\n",
421
- " <td>2023-07-01 12:00:00</td>\n",
422
- " <td>0.0688</td>\n",
423
- " <td>0.0114</td>\n",
424
- " <td>20.0</td>\n",
425
- " <td>17.0</td>\n",
426
- " <td>서울 중구 덕수궁길 15</td>\n",
427
- " <td>2023.0</td>\n",
428
- " <td>7.0</td>\n",
429
- " <td>1.0</td>\n",
430
- " <td>12.0</td>\n",
431
- " </tr>\n",
432
- " <tr>\n",
433
- " <th>12</th>\n",
434
- " <td>서울 중구</td>\n",
435
- " <td>도시대기</td>\n",
436
- " <td>111121</td>\n",
437
- " <td>중구</td>\n",
438
- " <td>2023-07-01 13:00:00</td>\n",
439
- " <td>0.0758</td>\n",
440
- " <td>0.0101</td>\n",
441
- " <td>23.0</td>\n",
442
- " <td>17.0</td>\n",
443
- " <td>서울 중구 덕수궁길 15</td>\n",
444
- " <td>2023.0</td>\n",
445
- " <td>7.0</td>\n",
446
- " <td>1.0</td>\n",
447
- " <td>13.0</td>\n",
448
- " </tr>\n",
449
- " <tr>\n",
450
- " <th>13</th>\n",
451
- " <td>서울 중구</td>\n",
452
- " <td>도시대기</td>\n",
453
- " <td>111121</td>\n",
454
- " <td>중구</td>\n",
455
- " <td>2023-07-01 14:00:00</td>\n",
456
- " <td>0.0743</td>\n",
457
- " <td>0.0093</td>\n",
458
- " <td>20.0</td>\n",
459
- " <td>17.0</td>\n",
460
- " <td>서울 중구 덕수궁길 15</td>\n",
461
- " <td>2023.0</td>\n",
462
- " <td>7.0</td>\n",
463
- " <td>1.0</td>\n",
464
- " <td>14.0</td>\n",
465
- " </tr>\n",
466
- " <tr>\n",
467
- " <th>14</th>\n",
468
- " <td>서울 중구</td>\n",
469
- " <td>도시대기</td>\n",
470
- " <td>111121</td>\n",
471
- " <td>중구</td>\n",
472
- " <td>2023-07-01 15:00:00</td>\n",
473
- " <td>0.0749</td>\n",
474
- " <td>0.0100</td>\n",
475
- " <td>19.0</td>\n",
476
- " <td>11.0</td>\n",
477
- " <td>서울 중구 덕수궁길 15</td>\n",
478
- " <td>2023.0</td>\n",
479
- " <td>7.0</td>\n",
480
- " <td>1.0</td>\n",
481
- " <td>15.0</td>\n",
482
- " </tr>\n",
483
- " <tr>\n",
484
- " <th>15</th>\n",
485
- " <td>서울 중구</td>\n",
486
- " <td>도시대기</td>\n",
487
- " <td>111121</td>\n",
488
- " <td>중구</td>\n",
489
- " <td>2023-07-01 16:00:00</td>\n",
490
- " <td>0.0716</td>\n",
491
- " <td>0.0092</td>\n",
492
- " <td>19.0</td>\n",
493
- " <td>15.0</td>\n",
494
- " <td>서울 중구 덕수궁길 15</td>\n",
495
- " <td>2023.0</td>\n",
496
- " <td>7.0</td>\n",
497
- " <td>1.0</td>\n",
498
- " <td>16.0</td>\n",
499
- " </tr>\n",
500
- " <tr>\n",
501
- " <th>16</th>\n",
502
- " <td>서울 중구</td>\n",
503
- " <td>도시대기</td>\n",
504
- " <td>111121</td>\n",
505
- " <td>중구</td>\n",
506
- " <td>2023-07-01 17:00:00</td>\n",
507
- " <td>0.0613</td>\n",
508
- " <td>0.0099</td>\n",
509
- " <td>18.0</td>\n",
510
- " <td>15.0</td>\n",
511
- " <td>서울 중구 덕수궁길 15</td>\n",
512
- " <td>2023.0</td>\n",
513
- " <td>7.0</td>\n",
514
- " <td>1.0</td>\n",
515
- " <td>17.0</td>\n",
516
- " </tr>\n",
517
- " <tr>\n",
518
- " <th>17</th>\n",
519
- " <td>서울 중구</td>\n",
520
- " <td>도시대기</td>\n",
521
- " <td>111121</td>\n",
522
- " <td>중구</td>\n",
523
- " <td>2023-07-01 18:00:00</td>\n",
524
- " <td>0.0496</td>\n",
525
- " <td>0.0098</td>\n",
526
- " <td>18.0</td>\n",
527
- " <td>14.0</td>\n",
528
- " <td>서울 중구 덕수궁길 15</td>\n",
529
- " <td>2023.0</td>\n",
530
- " <td>7.0</td>\n",
531
- " <td>1.0</td>\n",
532
- " <td>18.0</td>\n",
533
- " </tr>\n",
534
- " <tr>\n",
535
- " <th>18</th>\n",
536
- " <td>서울 중구</td>\n",
537
- " <td>도시대기</td>\n",
538
- " <td>111121</td>\n",
539
- " <td>중구</td>\n",
540
- " <td>2023-07-01 19:00:00</td>\n",
541
- " <td>0.0473</td>\n",
542
- " <td>0.0124</td>\n",
543
- " <td>17.0</td>\n",
544
- " <td>17.0</td>\n",
545
- " <td>서울 중구 덕수궁길 15</td>\n",
546
- " <td>2023.0</td>\n",
547
- " <td>7.0</td>\n",
548
- " <td>1.0</td>\n",
549
- " <td>19.0</td>\n",
550
- " </tr>\n",
551
- " <tr>\n",
552
- " <th>19</th>\n",
553
- " <td>서울 중구</td>\n",
554
- " <td>도시대기</td>\n",
555
- " <td>111121</td>\n",
556
- " <td>중구</td>\n",
557
- " <td>2023-07-01 20:00:00</td>\n",
558
- " <td>0.0498</td>\n",
559
- " <td>0.0170</td>\n",
560
- " <td>17.0</td>\n",
561
- " <td>15.0</td>\n",
562
- " <td>서울 중구 덕수궁길 15</td>\n",
563
- " <td>2023.0</td>\n",
564
- " <td>7.0</td>\n",
565
- " <td>1.0</td>\n",
566
- " <td>20.0</td>\n",
567
- " </tr>\n",
568
- " <tr>\n",
569
- " <th>20</th>\n",
570
- " <td>서울 중구</td>\n",
571
- " <td>도시대기</td>\n",
572
- " <td>111121</td>\n",
573
- " <td>중구</td>\n",
574
- " <td>2023-07-01 21:00:00</td>\n",
575
- " <td>0.0616</td>\n",
576
- " <td>0.0134</td>\n",
577
- " <td>23.0</td>\n",
578
- " <td>20.0</td>\n",
579
- " <td>서울 중구 덕수궁길 15</td>\n",
580
- " <td>2023.0</td>\n",
581
- " <td>7.0</td>\n",
582
- " <td>1.0</td>\n",
583
- " <td>21.0</td>\n",
584
- " </tr>\n",
585
- " <tr>\n",
586
- " <th>21</th>\n",
587
- " <td>서울 중구</td>\n",
588
- " <td>도시대기</td>\n",
589
- " <td>111121</td>\n",
590
- " <td>중구</td>\n",
591
- " <td>2023-07-01 22:00:00</td>\n",
592
- " <td>0.0543</td>\n",
593
- " <td>0.0109</td>\n",
594
- " <td>18.0</td>\n",
595
- " <td>16.0</td>\n",
596
- " <td>서울 중구 덕수궁길 15</td>\n",
597
- " <td>2023.0</td>\n",
598
- " <td>7.0</td>\n",
599
- " <td>1.0</td>\n",
600
- " <td>22.0</td>\n",
601
- " </tr>\n",
602
- " <tr>\n",
603
- " <th>22</th>\n",
604
- " <td>서울 중구</td>\n",
605
- " <td>도시대기</td>\n",
606
- " <td>111121</td>\n",
607
- " <td>중구</td>\n",
608
- " <td>2023-07-01 23:00:00</td>\n",
609
- " <td>0.0507</td>\n",
610
- " <td>0.0113</td>\n",
611
- " <td>17.0</td>\n",
612
- " <td>16.0</td>\n",
613
- " <td>서울 중구 덕수궁길 15</td>\n",
614
- " <td>2023.0</td>\n",
615
- " <td>7.0</td>\n",
616
- " <td>1.0</td>\n",
617
- " <td>23.0</td>\n",
618
- " </tr>\n",
619
- " <tr>\n",
620
- " <th>23</th>\n",
621
- " <td>서울 중구</td>\n",
622
- " <td>도시대기</td>\n",
623
- " <td>111121</td>\n",
624
- " <td>중구</td>\n",
625
- " <td>NaT</td>\n",
626
- " <td>0.0427</td>\n",
627
- " <td>0.0125</td>\n",
628
- " <td>17.0</td>\n",
629
- " <td>16.0</td>\n",
630
- " <td>서울 중구 덕수궁길 15</td>\n",
631
- " <td>NaN</td>\n",
632
- " <td>NaN</td>\n",
633
- " <td>NaN</td>\n",
634
- " <td>NaN</td>\n",
635
- " </tr>\n",
636
- " <tr>\n",
637
- " <th>24</th>\n",
638
- " <td>서울 중구</td>\n",
639
- " <td>도시대기</td>\n",
640
- " <td>111121</td>\n",
641
- " <td>중구</td>\n",
642
- " <td>2023-07-02 01:00:00</td>\n",
643
- " <td>0.0334</td>\n",
644
- " <td>0.0148</td>\n",
645
- " <td>21.0</td>\n",
646
- " <td>20.0</td>\n",
647
- " <td>서울 중구 덕수궁길 15</td>\n",
648
- " <td>2023.0</td>\n",
649
- " <td>7.0</td>\n",
650
- " <td>2.0</td>\n",
651
- " <td>1.0</td>\n",
652
- " </tr>\n",
653
- " <tr>\n",
654
- " <th>25</th>\n",
655
- " <td>서울 중구</td>\n",
656
- " <td>도시대기</td>\n",
657
- " <td>111121</td>\n",
658
- " <td>중구</td>\n",
659
- " <td>2023-07-02 02:00:00</td>\n",
660
- " <td>0.0337</td>\n",
661
- " <td>0.0133</td>\n",
662
- " <td>22.0</td>\n",
663
- " <td>18.0</td>\n",
664
- " <td>서울 중구 덕수궁길 15</td>\n",
665
- " <td>2023.0</td>\n",
666
- " <td>7.0</td>\n",
667
- " <td>2.0</td>\n",
668
- " <td>2.0</td>\n",
669
- " </tr>\n",
670
- " <tr>\n",
671
- " <th>26</th>\n",
672
- " <td>서울 중구</td>\n",
673
- " <td>도시대기</td>\n",
674
- " <td>111121</td>\n",
675
- " <td>중구</td>\n",
676
- " <td>2023-07-02 03:00:00</td>\n",
677
- " <td>0.0260</td>\n",
678
- " <td>0.0162</td>\n",
679
- " <td>25.0</td>\n",
680
- " <td>20.0</td>\n",
681
- " <td>서울 중구 덕수궁길 15</td>\n",
682
- " <td>2023.0</td>\n",
683
- " <td>7.0</td>\n",
684
- " <td>2.0</td>\n",
685
- " <td>3.0</td>\n",
686
- " </tr>\n",
687
- " <tr>\n",
688
- " <th>27</th>\n",
689
- " <td>서울 중구</td>\n",
690
- " <td>도시대기</td>\n",
691
- " <td>111121</td>\n",
692
- " <td>중구</td>\n",
693
- " <td>2023-07-02 04:00:00</td>\n",
694
- " <td>0.0195</td>\n",
695
- " <td>0.0179</td>\n",
696
- " <td>22.0</td>\n",
697
- " <td>18.0</td>\n",
698
- " <td>서울 중구 덕수궁길 15</td>\n",
699
- " <td>2023.0</td>\n",
700
- " <td>7.0</td>\n",
701
- " <td>2.0</td>\n",
702
- " <td>4.0</td>\n",
703
- " </tr>\n",
704
- " <tr>\n",
705
- " <th>28</th>\n",
706
- " <td>서울 중구</td>\n",
707
- " <td>도시대기</td>\n",
708
- " <td>111121</td>\n",
709
- " <td>중구</td>\n",
710
- " <td>2023-07-02 05:00:00</td>\n",
711
- " <td>0.0171</td>\n",
712
- " <td>0.0170</td>\n",
713
- " <td>19.0</td>\n",
714
- " <td>17.0</td>\n",
715
- " <td>서울 중구 덕수궁길 15</td>\n",
716
- " <td>2023.0</td>\n",
717
- " <td>7.0</td>\n",
718
- " <td>2.0</td>\n",
719
- " <td>5.0</td>\n",
720
- " </tr>\n",
721
- " <tr>\n",
722
- " <th>29</th>\n",
723
- " <td>서울 중구</td>\n",
724
- " <td>도시대기</td>\n",
725
- " <td>111121</td>\n",
726
- " <td>중구</td>\n",
727
- " <td>2023-07-02 06:00:00</td>\n",
728
- " <td>0.0181</td>\n",
729
- " <td>0.0145</td>\n",
730
- " <td>14.0</td>\n",
731
- " <td>10.0</td>\n",
732
- " <td>서울 중구 덕수궁길 15</td>\n",
733
- " <td>2023.0</td>\n",
734
- " <td>7.0</td>\n",
735
- " <td>2.0</td>\n",
736
- " <td>6.0</td>\n",
737
- " </tr>\n",
738
- " <tr>\n",
739
- " <th>30</th>\n",
740
- " <td>서울 중구</td>\n",
741
- " <td>도시대기</td>\n",
742
- " <td>111121</td>\n",
743
- " <td>중구</td>\n",
744
- " <td>2023-07-02 07:00:00</td>\n",
745
- " <td>0.0174</td>\n",
746
- " <td>0.0156</td>\n",
747
- " <td>11.0</td>\n",
748
- " <td>10.0</td>\n",
749
- " <td>서울 중구 덕수궁길 15</td>\n",
750
- " <td>2023.0</td>\n",
751
- " <td>7.0</td>\n",
752
- " <td>2.0</td>\n",
753
- " <td>7.0</td>\n",
754
- " </tr>\n",
755
- " <tr>\n",
756
- " <th>31</th>\n",
757
- " <td>서울 중구</td>\n",
758
- " <td>도시대기</td>\n",
759
- " <td>111121</td>\n",
760
- " <td>중구</td>\n",
761
- " <td>2023-07-02 08:00:00</td>\n",
762
- " <td>0.0213</td>\n",
763
- " <td>0.0147</td>\n",
764
- " <td>12.0</td>\n",
765
- " <td>9.0</td>\n",
766
- " <td>서울 중구 덕수궁길 15</td>\n",
767
- " <td>2023.0</td>\n",
768
- " <td>7.0</td>\n",
769
- " <td>2.0</td>\n",
770
- " <td>8.0</td>\n",
771
- " </tr>\n",
772
- " <tr>\n",
773
- " <th>32</th>\n",
774
- " <td>서울 중구</td>\n",
775
- " <td>도시대기</td>\n",
776
- " <td>111121</td>\n",
777
- " <td>중구</td>\n",
778
- " <td>2023-07-02 09:00:00</td>\n",
779
- " <td>0.0267</td>\n",
780
- " <td>0.0143</td>\n",
781
- " <td>11.0</td>\n",
782
- " <td>10.0</td>\n",
783
- " <td>서울 중구 덕수궁길 15</td>\n",
784
- " <td>2023.0</td>\n",
785
- " <td>7.0</td>\n",
786
- " <td>2.0</td>\n",
787
- " <td>9.0</td>\n",
788
- " </tr>\n",
789
- " <tr>\n",
790
- " <th>33</th>\n",
791
- " <td>서울 중구</td>\n",
792
- " <td>도시대기</td>\n",
793
- " <td>111121</td>\n",
794
- " <td>중구</td>\n",
795
- " <td>2023-07-02 10:00:00</td>\n",
796
- " <td>0.0289</td>\n",
797
- " <td>0.0155</td>\n",
798
- " <td>12.0</td>\n",
799
- " <td>9.0</td>\n",
800
- " <td>서울 중구 덕수궁길 15</td>\n",
801
- " <td>2023.0</td>\n",
802
- " <td>7.0</td>\n",
803
- " <td>2.0</td>\n",
804
- " <td>10.0</td>\n",
805
- " </tr>\n",
806
- " <tr>\n",
807
- " <th>34</th>\n",
808
- " <td>서울 중구</td>\n",
809
- " <td>도시대기</td>\n",
810
- " <td>111121</td>\n",
811
- " <td>중구</td>\n",
812
- " <td>2023-07-02 11:00:00</td>\n",
813
- " <td>0.0381</td>\n",
814
- " <td>0.0108</td>\n",
815
- " <td>13.0</td>\n",
816
- " <td>13.0</td>\n",
817
- " <td>서울 중구 덕수궁길 15</td>\n",
818
- " <td>2023.0</td>\n",
819
- " <td>7.0</td>\n",
820
- " <td>2.0</td>\n",
821
- " <td>11.0</td>\n",
822
- " </tr>\n",
823
- " <tr>\n",
824
- " <th>35</th>\n",
825
- " <td>서울 중구</td>\n",
826
- " <td>도시대기</td>\n",
827
- " <td>111121</td>\n",
828
- " <td>중구</td>\n",
829
- " <td>2023-07-02 12:00:00</td>\n",
830
- " <td>0.0441</td>\n",
831
- " <td>0.0079</td>\n",
832
- " <td>13.0</td>\n",
833
- " <td>12.0</td>\n",
834
- " <td>서울 중구 덕수궁길 15</td>\n",
835
- " <td>2023.0</td>\n",
836
- " <td>7.0</td>\n",
837
- " <td>2.0</td>\n",
838
- " <td>12.0</td>\n",
839
- " </tr>\n",
840
- " <tr>\n",
841
- " <th>36</th>\n",
842
- " <td>서울 중구</td>\n",
843
- " <td>도시대기</td>\n",
844
- " <td>111121</td>\n",
845
- " <td>중구</td>\n",
846
- " <td>2023-07-02 13:00:00</td>\n",
847
- " <td>0.0489</td>\n",
848
- " <td>0.0067</td>\n",
849
- " <td>8.0</td>\n",
850
- " <td>10.0</td>\n",
851
- " <td>서울 중구 덕수궁길 15</td>\n",
852
- " <td>2023.0</td>\n",
853
- " <td>7.0</td>\n",
854
- " <td>2.0</td>\n",
855
- " <td>13.0</td>\n",
856
- " </tr>\n",
857
- " <tr>\n",
858
- " <th>37</th>\n",
859
- " <td>서울 중구</td>\n",
860
- " <td>도시대기</td>\n",
861
- " <td>111121</td>\n",
862
- " <td>중구</td>\n",
863
- " <td>2023-07-02 14:00:00</td>\n",
864
- " <td>0.0498</td>\n",
865
- " <td>0.0072</td>\n",
866
- " <td>11.0</td>\n",
867
- " <td>10.0</td>\n",
868
- " <td>서울 중구 덕수궁길 15</td>\n",
869
- " <td>2023.0</td>\n",
870
- " <td>7.0</td>\n",
871
- " <td>2.0</td>\n",
872
- " <td>14.0</td>\n",
873
- " </tr>\n",
874
- " <tr>\n",
875
- " <th>38</th>\n",
876
- " <td>서울 중구</td>\n",
877
- " <td>도시대기</td>\n",
878
- " <td>111121</td>\n",
879
- " <td>중구</td>\n",
880
- " <td>2023-07-02 15:00:00</td>\n",
881
- " <td>0.0459</td>\n",
882
- " <td>0.0073</td>\n",
883
- " <td>14.0</td>\n",
884
- " <td>12.0</td>\n",
885
- " <td>서울 중구 덕수궁길 15</td>\n",
886
- " <td>2023.0</td>\n",
887
- " <td>7.0</td>\n",
888
- " <td>2.0</td>\n",
889
- " <td>15.0</td>\n",
890
- " </tr>\n",
891
- " <tr>\n",
892
- " <th>39</th>\n",
893
- " <td>서울 중구</td>\n",
894
- " <td>도시대기</td>\n",
895
- " <td>111121</td>\n",
896
- " <td>중구</td>\n",
897
- " <td>2023-07-02 16:00:00</td>\n",
898
- " <td>0.0474</td>\n",
899
- " <td>0.0079</td>\n",
900
- " <td>12.0</td>\n",
901
- " <td>11.0</td>\n",
902
- " <td>서울 중구 덕수궁길 15</td>\n",
903
- " <td>2023.0</td>\n",
904
- " <td>7.0</td>\n",
905
- " <td>2.0</td>\n",
906
- " <td>16.0</td>\n",
907
- " </tr>\n",
908
- " </tbody>\n",
909
- "</table>\n",
910
- "</div>"
911
- ],
912
- "text/plain": [
913
- " 지역 망 측정소코드 측정소명 측정일시 O3 NO2 PM10 PM25 \\\n",
914
- "0 서울 중구 도시대기 111121 중구 2023-07-01 01:00:00 0.0249 0.0188 21.0 19.0 \n",
915
- "1 서울 중구 도시대기 111121 중구 2023-07-01 02:00:00 0.0263 0.0163 18.0 15.0 \n",
916
- "2 서울 중구 도시대기 111121 중구 2023-07-01 03:00:00 0.0218 0.0192 24.0 21.0 \n",
917
- "3 서울 중구 도시대기 111121 중구 2023-07-01 04:00:00 0.0131 0.0214 25.0 19.0 \n",
918
- "4 서울 중구 도시대기 111121 중구 2023-07-01 05:00:00 0.0131 0.0160 25.0 21.0 \n",
919
- "5 서울 중구 도시대기 111121 중구 2023-07-01 06:00:00 0.0115 0.0196 23.0 18.0 \n",
920
- "6 서울 중구 도시대기 111121 중구 2023-07-01 07:00:00 0.0094 0.0230 26.0 21.0 \n",
921
- "7 서울 중구 도시대기 111121 중구 2023-07-01 08:00:00 0.0222 0.0175 26.0 20.0 \n",
922
- "8 서울 중구 도시대기 111121 중구 2023-07-01 09:00:00 0.0396 0.0153 27.0 20.0 \n",
923
- "9 서울 중구 도시대기 111121 중구 2023-07-01 10:00:00 0.0530 0.0105 19.0 16.0 \n",
924
- "10 서울 중구 도시대기 111121 중구 2023-07-01 11:00:00 0.0607 0.0090 20.0 20.0 \n",
925
- "11 서울 중구 도시대기 111121 중구 2023-07-01 12:00:00 0.0688 0.0114 20.0 17.0 \n",
926
- "12 서울 중구 도시대기 111121 중구 2023-07-01 13:00:00 0.0758 0.0101 23.0 17.0 \n",
927
- "13 서울 중구 도시대기 111121 중구 2023-07-01 14:00:00 0.0743 0.0093 20.0 17.0 \n",
928
- "14 서울 중구 도시대기 111121 중구 2023-07-01 15:00:00 0.0749 0.0100 19.0 11.0 \n",
929
- "15 서울 중구 도시대기 111121 중구 2023-07-01 16:00:00 0.0716 0.0092 19.0 15.0 \n",
930
- "16 서울 중구 도시대기 111121 중구 2023-07-01 17:00:00 0.0613 0.0099 18.0 15.0 \n",
931
- "17 서울 중구 도시대기 111121 중구 2023-07-01 18:00:00 0.0496 0.0098 18.0 14.0 \n",
932
- "18 서울 중구 도시대기 111121 중구 2023-07-01 19:00:00 0.0473 0.0124 17.0 17.0 \n",
933
- "19 서울 중구 도시대기 111121 중구 2023-07-01 20:00:00 0.0498 0.0170 17.0 15.0 \n",
934
- "20 서울 중구 도시대기 111121 중구 2023-07-01 21:00:00 0.0616 0.0134 23.0 20.0 \n",
935
- "21 서울 중구 도시대기 111121 중구 2023-07-01 22:00:00 0.0543 0.0109 18.0 16.0 \n",
936
- "22 서울 중구 도시대기 111121 중구 2023-07-01 23:00:00 0.0507 0.0113 17.0 16.0 \n",
937
- "23 서울 중구 도시대기 111121 중구 NaT 0.0427 0.0125 17.0 16.0 \n",
938
- "24 서울 중구 도시대기 111121 중구 2023-07-02 01:00:00 0.0334 0.0148 21.0 20.0 \n",
939
- "25 서울 중구 도시대기 111121 중구 2023-07-02 02:00:00 0.0337 0.0133 22.0 18.0 \n",
940
- "26 서울 중구 도시대기 111121 중구 2023-07-02 03:00:00 0.0260 0.0162 25.0 20.0 \n",
941
- "27 서울 중구 도시대기 111121 중구 2023-07-02 04:00:00 0.0195 0.0179 22.0 18.0 \n",
942
- "28 서울 중구 도시대기 111121 중구 2023-07-02 05:00:00 0.0171 0.0170 19.0 17.0 \n",
943
- "29 서울 중구 도시대기 111121 중구 2023-07-02 06:00:00 0.0181 0.0145 14.0 10.0 \n",
944
- "30 서울 중구 도시대기 111121 중구 2023-07-02 07:00:00 0.0174 0.0156 11.0 10.0 \n",
945
- "31 서울 중구 도시대기 111121 중구 2023-07-02 08:00:00 0.0213 0.0147 12.0 9.0 \n",
946
- "32 서울 중구 도시대기 111121 중구 2023-07-02 09:00:00 0.0267 0.0143 11.0 10.0 \n",
947
- "33 서울 중구 도시대기 111121 중구 2023-07-02 10:00:00 0.0289 0.0155 12.0 9.0 \n",
948
- "34 서울 중구 도시대기 111121 중구 2023-07-02 11:00:00 0.0381 0.0108 13.0 13.0 \n",
949
- "35 서울 중구 도시대기 111121 중구 2023-07-02 12:00:00 0.0441 0.0079 13.0 12.0 \n",
950
- "36 서울 중구 도시대기 111121 중구 2023-07-02 13:00:00 0.0489 0.0067 8.0 10.0 \n",
951
- "37 서울 중구 도시대기 111121 중구 2023-07-02 14:00:00 0.0498 0.0072 11.0 10.0 \n",
952
- "38 서울 중구 도시대기 111121 중구 2023-07-02 15:00:00 0.0459 0.0073 14.0 12.0 \n",
953
- "39 서울 중구 도시대기 111121 중구 2023-07-02 16:00:00 0.0474 0.0079 12.0 11.0 \n",
954
- "\n",
955
- " 주소 year month day hour \n",
956
- "0 서울 중구 덕수궁길 15 2023.0 7.0 1.0 1.0 \n",
957
- "1 서울 중구 덕수궁길 15 2023.0 7.0 1.0 2.0 \n",
958
- "2 서울 중구 덕수궁길 15 2023.0 7.0 1.0 3.0 \n",
959
- "3 서울 중구 덕수궁길 15 2023.0 7.0 1.0 4.0 \n",
960
- "4 서울 중구 덕수궁길 15 2023.0 7.0 1.0 5.0 \n",
961
- "5 서울 중구 덕수궁길 15 2023.0 7.0 1.0 6.0 \n",
962
- "6 서울 중구 덕수궁길 15 2023.0 7.0 1.0 7.0 \n",
963
- "7 서울 중구 덕수궁길 15 2023.0 7.0 1.0 8.0 \n",
964
- "8 서울 중구 덕수궁길 15 2023.0 7.0 1.0 9.0 \n",
965
- "9 서울 중구 덕수궁길 15 2023.0 7.0 1.0 10.0 \n",
966
- "10 서울 중구 덕수궁길 15 2023.0 7.0 1.0 11.0 \n",
967
- "11 서울 중구 덕수궁길 15 2023.0 7.0 1.0 12.0 \n",
968
- "12 서울 중구 덕수궁길 15 2023.0 7.0 1.0 13.0 \n",
969
- "13 서울 중구 덕수궁길 15 2023.0 7.0 1.0 14.0 \n",
970
- "14 서울 중구 덕수궁길 15 2023.0 7.0 1.0 15.0 \n",
971
- "15 서울 중구 덕수궁길 15 2023.0 7.0 1.0 16.0 \n",
972
- "16 서울 중구 덕수궁길 15 2023.0 7.0 1.0 17.0 \n",
973
- "17 서울 중구 덕수궁길 15 2023.0 7.0 1.0 18.0 \n",
974
- "18 서울 중구 덕수궁길 15 2023.0 7.0 1.0 19.0 \n",
975
- "19 서울 중구 덕수궁길 15 2023.0 7.0 1.0 20.0 \n",
976
- "20 서울 중구 덕수궁길 15 2023.0 7.0 1.0 21.0 \n",
977
- "21 서울 중구 덕수궁길 15 2023.0 7.0 1.0 22.0 \n",
978
- "22 서울 중구 덕수궁길 15 2023.0 7.0 1.0 23.0 \n",
979
- "23 서울 중구 덕수궁길 15 NaN NaN NaN NaN \n",
980
- "24 서울 중구 덕수궁길 15 2023.0 7.0 2.0 1.0 \n",
981
- "25 서울 중구 덕수궁길 15 2023.0 7.0 2.0 2.0 \n",
982
- "26 서울 중구 덕수궁길 15 2023.0 7.0 2.0 3.0 \n",
983
- "27 서울 중구 덕수궁길 15 2023.0 7.0 2.0 4.0 \n",
984
- "28 서울 중구 덕수궁길 15 2023.0 7.0 2.0 5.0 \n",
985
- "29 서울 중구 덕수궁길 15 2023.0 7.0 2.0 6.0 \n",
986
- "30 서울 중구 덕수궁길 15 2023.0 7.0 2.0 7.0 \n",
987
- "31 서울 중구 덕수궁길 15 2023.0 7.0 2.0 8.0 \n",
988
- "32 서울 중구 덕수궁길 15 2023.0 7.0 2.0 9.0 \n",
989
- "33 서울 중구 덕수궁길 15 2023.0 7.0 2.0 10.0 \n",
990
- "34 서울 중구 덕수궁길 15 2023.0 7.0 2.0 11.0 \n",
991
- "35 서울 중구 덕수궁길 15 2023.0 7.0 2.0 12.0 \n",
992
- "36 서울 중구 덕수궁길 15 2023.0 7.0 2.0 13.0 \n",
993
- "37 서울 중구 덕수궁길 15 2023.0 7.0 2.0 14.0 \n",
994
- "38 서울 중구 덕수궁길 15 2023.0 7.0 2.0 15.0 \n",
995
- "39 서울 중구 덕수궁길 15 2023.0 7.0 2.0 16.0 "
996
- ]
997
- },
998
- "execution_count": 6,
999
- "metadata": {},
1000
- "output_type": "execute_result"
1001
- }
1002
- ],
1003
- "source": [
1004
- "data.head(40)"
1005
- ]
1006
- }
1007
- ],
1008
- "metadata": {
1009
- "kernelspec": {
1010
- "display_name": "Python 3 (ipykernel)",
1011
- "language": "python",
1012
- "name": "python3"
1013
- },
1014
- "language_info": {
1015
- "codemirror_mode": {
1016
- "name": "ipython",
1017
- "version": 3
1018
- },
1019
- "file_extension": ".py",
1020
- "mimetype": "text/x-python",
1021
- "name": "python",
1022
- "nbconvert_exporter": "python",
1023
- "pygments_lexer": "ipython3",
1024
- "version": "3.8.13"
1025
- }
1026
- },
1027
- "nbformat": 4,
1028
- "nbformat_minor": 4
1029
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/1.data_merge.ipynb DELETED
The diff for this file is too large to render. See raw diff
 
Analysis_code/2.eda_preproccesing.ipynb DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:f8fe49fd26e48bcada89076a0b3c8ffe45e3d2e8407fe953b1558df9bfcfcddb
3
- size 41518580
 
 
 
 
Analysis_code/3.oversampling.ipynb DELETED
The diff for this file is too large to render. See raw diff
 
Analysis_code/__pycache__/code.cpython-312.pyc DELETED
Binary file (13.8 kB)
 
Analysis_code/__pycache__/deepgbm.cpython-312.pyc DELETED
Binary file (2.67 kB)
 
Analysis_code/__pycache__/deepgbm.cpython-38.pyc DELETED
Binary file (1.88 kB)
 
Analysis_code/__pycache__/deepgbm.cpython-39.pyc DELETED
Binary file (1.82 kB)
 
Analysis_code/__pycache__/ft_transformer.cpython-312.pyc DELETED
Binary file (2.71 kB)
 
Analysis_code/__pycache__/ft_transformer.cpython-38.pyc DELETED
Binary file (1.99 kB)
 
Analysis_code/__pycache__/ft_transformer.cpython-39.pyc DELETED
Binary file (1.93 kB)
 
Analysis_code/__pycache__/resnet_like.cpython-312.pyc DELETED
Binary file (2.26 kB)
 
Analysis_code/__pycache__/resnet_like.cpython-38.pyc DELETED
Binary file (1.54 kB)
 
Analysis_code/__pycache__/resnet_like.cpython-39.pyc DELETED
Binary file (1.48 kB)
 
Analysis_code/best_deepgbm_model.pth DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:471ce03d818bdefc9631121537afd4b771d85d872f72d74634e5cee824de2b62
3
- size 988522
 
 
 
 
Analysis_code/best_model_f1.pth DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:12c6e86e039528b07c4cc6e90e3033da7b02d67c33539cbb81d280da281f46c3
3
- size 8999933
 
 
 
 
Analysis_code/deepgbm.py DELETED
@@ -1,47 +0,0 @@
1
- import torch
2
- import torch.nn as nn
3
- import torch.nn.functional as F
4
-
5
- class DeepGBM(nn.Module):
6
- def __init__(self, num_features, cat_features, num_classes, d_main=128, d_hidden=64, n_blocks=4, dropout=0.2):
7
- super(DeepGBM, self).__init__()
8
-
9
- self.num_classes = num_classes
10
-
11
- # 연속형 변수 처리 (Linear)
12
- self.num_linear = nn.Linear(num_features, d_main)
13
-
14
- # 범주형 변수 처리 (Embedding)
15
- self.cat_embedding = nn.ModuleList([
16
- nn.Embedding(cat_size, d_main) for cat_size in cat_features
17
- ])
18
-
19
- # ResNet-like 블록
20
- self.blocks = nn.ModuleList([
21
- nn.Sequential(
22
- nn.Linear(d_main, d_hidden),
23
- nn.ReLU(),
24
- nn.Dropout(dropout),
25
- nn.Linear(d_hidden, d_main),
26
- nn.ReLU()
27
- ) for _ in range(n_blocks)
28
- ])
29
-
30
- if num_classes == 2:
31
- self.output_layer = nn.Linear(d_main, 1) # Binary classification
32
- elif num_classes > 2:
33
- self.output_layer = nn.Linear(d_main, num_classes) # Multi classification
34
-
35
- def forward(self, x_num, x_cat): # 두 개의 입력을 받음
36
- x_num = self.num_linear(x_num)
37
-
38
- # 범주형 변수를 임베딩 후 합산
39
- x_cat = [embed(x_cat[:, i]) for i, embed in enumerate(self.cat_embedding)]
40
- x_cat = torch.stack(x_cat, dim=1).sum(dim=1)
41
- x = x_num + x_cat # 연속형 + 범주형 결합
42
-
43
- for block in self.blocks:
44
- x = x + block(x) # Residual connection
45
- x = self.output_layer(x)
46
-
47
- return x
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/deeplearning_model_binary.ipynb DELETED
The diff for this file is too large to render. See raw diff
 
Analysis_code/deeplearning_model_multi.ipynb DELETED
The diff for this file is too large to render. See raw diff
 
Analysis_code/final_test/final.ipynb DELETED
@@ -1,1143 +0,0 @@
1
- {
2
- "cells": [
3
- {
4
- "cell_type": "code",
5
- "execution_count": 1,
6
- "metadata": {},
7
- "outputs": [],
8
- "source": [
9
- "import pandas as pd\n",
10
- "import numpy as np\n",
11
- "from sklearn.preprocessing import StandardScaler, LabelEncoder\n",
12
- "import torch\n",
13
- "from torch.utils.data import DataLoader, TensorDataset\n",
14
- "import random\n",
15
- "from collections import Counter\n",
16
- "import sys\n",
17
- "sys.path.append('../../../../../../../../mnt/workspace/LightGBM/python-package')\n",
18
- "from lightgbm import LGBMClassifier\n",
19
- "import numpy as np\n",
20
- "from sklearn.model_selection import train_test_split\n",
21
- "from sklearn.inspection import permutation_importance\n",
22
- "from sklearn.metrics import confusion_matrix,accuracy_score, recall_score, precision_score\n",
23
- "from sklearn.model_selection import StratifiedKFold\n",
24
- "from xgboost import XGBClassifier\n",
25
- "from warnings import filterwarnings\n",
26
- "filterwarnings('ignore')\n",
27
- "import sys\n",
28
- "sys.path.append('../')\n",
29
- "import torch\n",
30
- "import torch.nn as nn\n",
31
- "import torch.optim as optim\n",
32
- "import optuna\n",
33
- "import pandas as pd\n",
34
- "import numpy as np\n",
35
- "import random\n",
36
- "from ft_transformer import FTTransformer\n",
37
- "from resnet_like import ResNetLike\n",
38
- "from deepgbm import DeepGBM\n",
39
- "from pytorch_tabnet.tab_model import TabNetClassifier\n",
40
- "from sklearn.metrics import accuracy_score, f1_score, roc_auc_score, confusion_matrix"
41
- ]
42
- },
43
- {
44
- "cell_type": "code",
45
- "execution_count": 2,
46
- "metadata": {},
47
- "outputs": [],
48
- "source": [
49
- "# Python 및 Numpy 시드 고정\n",
50
- "seed = 42\n",
51
- "random.seed(seed)\n",
52
- "np.random.seed(seed)\n",
53
- "\n",
54
- "# PyTorch 시드 고정\n",
55
- "torch.manual_seed(seed)\n",
56
- "torch.cuda.manual_seed(seed)\n",
57
- "torch.cuda.manual_seed_all(seed) # Multi-GPU 환경에서 동일한 시드 적용\n",
58
- "\n",
59
- "# PyTorch 연산의 결정적 모드 설정\n",
60
- "torch.backends.cudnn.deterministic = True # 실행마다 동일한 결과를 보장\n",
61
- "torch.backends.cudnn.benchmark = True # 성능 최적화를 활성화 (가능한 한 빠른 연산 수행)\n",
62
- "\n",
63
- "# 전처리 함수\n",
64
- "def preprocessing(df):\n",
65
- " df = df[df.columns].copy()\n",
66
- " df.loc[df['wind_dir']=='정온', 'wind_dir'] = \"0\"\n",
67
- " df['wind_dir'] = df['wind_dir'].astype('int')\n",
68
- " df['lm_cloudcover'] = df['lm_cloudcover'].astype('int')\n",
69
- " df['cloudcover'] = df['cloudcover'].astype('int')\n",
70
- " return df\n",
71
- "\n",
72
- "# 데이터셋 준비 함수\n",
73
- "def prepare_dataset(region, data_sample='pure', target='multi', fold=3):\n",
74
- "\n",
75
- " # 데이터 경로 지정\n",
76
- " dat_path = f\"../../data/data_for_modeling/{region}_train.csv\"\n",
77
- " if data_sample == 'pure':\n",
78
- " train_path = dat_path\n",
79
- " else:\n",
80
- " train_path = f'../../data/data_oversampled/{data_sample}/{data_sample}_{fold}_{region}.csv'\n",
81
- " train_path = f'../../data/data_oversampled/{data_sample}/{data_sample}_{fold}_{region}.csv'\n",
82
- " test_path = f\"../../data/data_for_modeling/{region}_test.csv\"\n",
83
- " drop_col = ['binary_class','multi_class','visi','year']\n",
84
- " target_col = f'{target}_class'\n",
85
- " \n",
86
- " # 데이터 로드\n",
87
- " region_dat = preprocessing(pd.read_csv(dat_path, index_col=0))\n",
88
- " if data_sample == 'pure':\n",
89
- " region_train = region_dat.loc[~region_dat['year'].isin([2021-fold]), :]\n",
90
- " else:\n",
91
- " region_train = preprocessing(pd.read_csv(train_path))\n",
92
- " region_val = region_dat.loc[region_dat['year'].isin([2021-fold]), :]\n",
93
- " region_test = preprocessing(pd.read_csv(test_path))\n",
94
- "\n",
95
- " # 컬럼 정렬 (일관성 유지)\n",
96
- " common_columns = region_train.columns.to_list()\n",
97
- " train_data = region_train[common_columns]\n",
98
- " val_data = region_val[common_columns]\n",
99
- " test_data = region_test[common_columns]\n",
100
- "\n",
101
- " # 설명변수 & 타겟 분리\n",
102
- " X_train = train_data.drop(columns=drop_col)\n",
103
- " y_train = train_data[target_col]\n",
104
- " X_val = val_data.drop(columns=drop_col)\n",
105
- " y_val = val_data[target_col]\n",
106
- " X_test = test_data.drop(columns=drop_col)\n",
107
- " y_test = test_data[target_col]\n",
108
- "\n",
109
- " # 범주형 & 연속형 변수 분리\n",
110
- " categorical_cols = X_train.select_dtypes(include=['object', 'category', 'int64']).columns\n",
111
- " numerical_cols = X_train.select_dtypes(include=['float64']).columns\n",
112
- "\n",
113
- " # 범주형 변수 Label Encoding\n",
114
- " label_encoders = {}\n",
115
- " for col in categorical_cols:\n",
116
- " le = LabelEncoder()\n",
117
- " le.fit(X_train[col]) # Train 데이터 기준으로 학���\n",
118
- " label_encoders[col] = le\n",
119
- "\n",
120
- " # 변환 적용\n",
121
- " for col in categorical_cols:\n",
122
- " X_train[col] = label_encoders[col].transform(X_train[col])\n",
123
- " X_val[col] = label_encoders[col].transform(X_val[col])\n",
124
- " X_test[col] = label_encoders[col].transform(X_test[col])\n",
125
- "\n",
126
- " # 연속형 변수 Standard Scaling\n",
127
- " scaler = StandardScaler()\n",
128
- " scaler.fit(X_train[numerical_cols]) # Train 데이터 기준으로 학습\n",
129
- "\n",
130
- " # 변환 적용\n",
131
- " X_train[numerical_cols] = scaler.transform(X_train[numerical_cols])\n",
132
- " X_val[numerical_cols] = scaler.transform(X_val[numerical_cols])\n",
133
- " X_test[numerical_cols] = scaler.transform(X_test[numerical_cols])\n",
134
- "\n",
135
- " return X_train, X_val, X_test, y_train, y_val, y_test, categorical_cols, numerical_cols\n",
136
- "\n",
137
- "# 데이터 변환 및 dataloader 생성 함수\n",
138
- "def prepare_dataloader(region, data_sample='pure', target='multi', fold=3, random_state=None):\n",
139
- "\n",
140
- " # 데이터 경로 지정\n",
141
- " dat_path = f\"../../data/data_for_modeling/{region}_train.csv\"\n",
142
- " if data_sample == 'pure':\n",
143
- " train_path = dat_path\n",
144
- " else:\n",
145
- " train_path = f'../../data/data_oversampled/{data_sample}/{data_sample}_{fold}_{region}.csv'\n",
146
- " train_path = f'../../data/data_oversampled/{data_sample}/{data_sample}_{fold}_{region}.csv'\n",
147
- " test_path = f\"../../data/data_for_modeling/{region}_test.csv\"\n",
148
- " drop_col = ['binary_class','multi_class','visi','year']\n",
149
- " target_col = f'{target}_class'\n",
150
- " \n",
151
- " # 데이터 로드\n",
152
- " region_dat = preprocessing(pd.read_csv(dat_path, index_col=0))\n",
153
- " if data_sample == 'pure':\n",
154
- " region_train = region_dat.loc[~region_dat['year'].isin([2021-fold]), :]\n",
155
- " else:\n",
156
- " region_train = preprocessing(pd.read_csv(train_path))\n",
157
- " region_val = region_dat.loc[region_dat['year'].isin([2021-fold]), :]\n",
158
- " region_test = preprocessing(pd.read_csv(test_path))\n",
159
- "\n",
160
- " # 컬럼 정렬 (일관성 유지)\n",
161
- " common_columns = region_train.columns.to_list()\n",
162
- " train_data = region_train[common_columns]\n",
163
- " val_data = region_val[common_columns]\n",
164
- " test_data = region_test[common_columns]\n",
165
- "\n",
166
- " # 설명변수 & 타겟 분리\n",
167
- " X_train = train_data.drop(columns=drop_col)\n",
168
- " y_train = train_data[target_col]\n",
169
- " X_val = val_data.drop(columns=drop_col)\n",
170
- " y_val = val_data[target_col]\n",
171
- " X_test = test_data.drop(columns=drop_col)\n",
172
- " y_test = test_data[target_col]\n",
173
- "\n",
174
- " # 범주형 & 연속형 변수 분리\n",
175
- " categorical_cols = X_train.select_dtypes(include=['object', 'category', 'int64']).columns\n",
176
- " numerical_cols = X_train.select_dtypes(include=['float64']).columns\n",
177
- "\n",
178
- " # 범주형 변수 Label Encoding\n",
179
- " label_encoders = {}\n",
180
- " for col in categorical_cols:\n",
181
- " le = LabelEncoder()\n",
182
- " le.fit(X_train[col]) # Train 데이터 기준으로 학습\n",
183
- " label_encoders[col] = le\n",
184
- "\n",
185
- " # 변환 적용\n",
186
- " for col in categorical_cols:\n",
187
- " X_train[col] = label_encoders[col].transform(X_train[col])\n",
188
- " X_val[col] = label_encoders[col].transform(X_val[col])\n",
189
- " X_test[col] = label_encoders[col].transform(X_test[col])\n",
190
- "\n",
191
- " # 연속형 변수 Standard Scaling\n",
192
- " scaler = StandardScaler()\n",
193
- " scaler.fit(X_train[numerical_cols]) # Train 데이터 기준으로 학습\n",
194
- "\n",
195
- " # 변환 적용\n",
196
- " X_train[numerical_cols] = scaler.transform(X_train[numerical_cols])\n",
197
- " X_val[numerical_cols] = scaler.transform(X_val[numerical_cols])\n",
198
- " X_test[numerical_cols] = scaler.transform(X_test[numerical_cols])\n",
199
- "\n",
200
- " # 연속형 변수와 범주형 변수 분리\n",
201
- " X_train_num = torch.tensor(X_train[numerical_cols].values, dtype=torch.float32)\n",
202
- " X_train_cat = torch.tensor(X_train[categorical_cols].values, dtype=torch.long)\n",
203
- "\n",
204
- " X_val_num = torch.tensor(X_val[numerical_cols].values, dtype=torch.float32)\n",
205
- " X_val_cat = torch.tensor(X_val[categorical_cols].values, dtype=torch.long)\n",
206
- "\n",
207
- " X_test_num = torch.tensor(X_test[numerical_cols].values, dtype=torch.float32)\n",
208
- " X_test_cat = torch.tensor(X_test[categorical_cols].values, dtype=torch.long)\n",
209
- "\n",
210
- " # 레이블 변환\n",
211
- " if target == \"binary\":\n",
212
- " y_train_tensor = torch.tensor(y_train.values, dtype=torch.float32) # 이진 분류 → float32\n",
213
- " y_val_tensor = torch.tensor(y_val.values, dtype=torch.float32)\n",
214
- " y_test_tensor = torch.tensor(y_test.values, dtype=torch.float32)\n",
215
- " elif target == \"multi\":\n",
216
- " y_train_tensor = torch.tensor(y_train.values, dtype=torch.long) # 다중 분류 → long\n",
217
- " y_val_tensor = torch.tensor(y_val.values, dtype=torch.long)\n",
218
- " y_test_tensor = torch.tensor(y_test.values, dtype=torch.long)\n",
219
- " else:\n",
220
- " raise ValueError(\"target must be 'binary' or 'multi'\")\n",
221
- "\n",
222
- " # TensorDataset 생성\n",
223
- " train_dataset = TensorDataset(X_train_num, X_train_cat, y_train_tensor)\n",
224
- " val_dataset = TensorDataset(X_val_num, X_val_cat, y_val_tensor)\n",
225
- " test_dataset = TensorDataset(X_test_num, X_test_cat, y_test_tensor)\n",
226
- "\n",
227
- " # DataLoader 생성\n",
228
- " if random_state == None:\n",
229
- " train_loader = DataLoader(train_dataset, batch_size=64, shuffle=True)\n",
230
- " else:\n",
231
- " train_loader = DataLoader(train_dataset, batch_size=64, shuffle=True, generator=torch.Generator().manual_seed(random_state))\n",
232
- " val_loader = DataLoader(val_dataset, batch_size=64, shuffle=False)\n",
233
- " test_loader = DataLoader(test_dataset, batch_size=64, shuffle=False)\n",
234
- " \n",
235
- " return X_train, categorical_cols, numerical_cols, train_loader, val_loader, test_loader"
236
- ]
237
- },
238
- {
239
- "cell_type": "code",
240
- "execution_count": 3,
241
- "metadata": {},
242
- "outputs": [],
243
- "source": [
244
- "import os\n",
245
- "import torch\n",
246
- "# 디바이스 설정 (CUDA 사용 가능하면 GPU로, 아니면 CPU로)\n",
247
- "import glob\n",
248
- "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
249
- "\n",
250
- "def calculate_csi(Y_test, pred):\n",
251
- "\n",
252
- " cm = confusion_matrix(Y_test, pred) # 변수 이름을 cm으로 변경\n",
253
- " # 혼동 행렬에서 H, F, M 추출\n",
254
- " H = (cm[0, 0] + cm[1, 1])\n",
255
- " \n",
256
- " F = (cm[1, 0] + cm[2, 0] +\n",
257
- " cm[0, 1] + cm[2, 1])\n",
258
- "\n",
259
- " M = (cm[0, 2] + cm[1, 2])\n",
260
- " \n",
261
- " # CSI 계산\n",
262
- " CSI = H / (H + F + M + 1e-10)\n",
263
- " return CSI\n",
264
- "\n",
265
- "def csi_metric(y_true, pred):\n",
266
- " y_pred_binary = np.argmax(pred, axis=1)\n",
267
- " score = calculate_csi(y_true, y_pred_binary)\n",
268
- " return 'CSI', score, True # higher_better=True\n",
269
- "\n",
270
- "\n",
271
- "def eval_metric_csi(y_true, pred_prob):\n",
272
- "\n",
273
- " pred = np.argmax(pred_prob, axis=1)\n",
274
- " y_true = y_true\n",
275
- " y_pred = pred\n",
276
- " csi = calculate_csi(y_true, y_pred)\n",
277
- " return -1*csi\n",
278
- "\n",
279
- "\n",
280
- "from sklearn.metrics import matthews_corrcoef, accuracy_score\n",
281
- "\n",
282
- "def multiclass_mcc(y_val, y_pred):\n",
283
- " \"\"\"\n",
284
- " 다중 분류에서도 sklearn의 matthews_corrcoef를 그대로 사용할 수 있음.\n",
285
- " \"\"\"\n",
286
- " return matthews_corrcoef(y_val, y_pred)"
287
- ]
288
- },
289
- {
290
- "cell_type": "code",
291
- "execution_count": 4,
292
- "metadata": {},
293
- "outputs": [],
294
- "source": [
295
- "import os\n",
296
- "import torch\n",
297
- "# 디바이스 설정 (CUDA 사용 가능하면 GPU로, 아니면 CPU로)\n",
298
- "import glob\n",
299
- "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
300
- "import warnings\n",
301
- "warnings.filterwarnings('ignore')\n",
302
- "\n",
303
- "def calculate_csi(Y_test, pred):\n",
304
- "\n",
305
- " cm = confusion_matrix(Y_test, pred) # 변수 이름을 cm으로 변경\n",
306
- " # 혼동 행렬에서 H, F, M 추출\n",
307
- " H = (cm[0, 0] + cm[1, 1])\n",
308
- " \n",
309
- " F = (cm[1, 0] + cm[2, 0] +\n",
310
- " cm[0, 1] + cm[2, 1])\n",
311
- "\n",
312
- " M = (cm[0, 2] + cm[1, 2])\n",
313
- " \n",
314
- " # CSI 계산\n",
315
- " CSI = H / (H + F + M + 1e-10)\n",
316
- " return CSI\n",
317
- "\n",
318
- "# Soft Voting 앙상블\n",
319
- "def get_proba(region, model_choose, data_sample, fold, target='multi'):\n",
320
- " _, _, _, _,val_loader , test_loader = prepare_dataloader(region=region, data_sample=data_sample, target=target,fold=fold ,random_state=120)\n",
321
- "\n",
322
- " folder_path = f'../save_model/{model_choose}/{data_sample}'\n",
323
- " model_paths = [path for path in glob.glob(f'{folder_path}/*.pth') if f'{region}' in path]\n",
324
- "\n",
325
- " model = torch.load(model_paths[fold-1], weights_only=False).to(device)\n",
326
- " model.eval()\n",
327
- "\n",
328
- " test_preds = []\n",
329
- "\n",
330
- "\n",
331
- " with torch.no_grad():\n",
332
- " for x_num_batch, x_cat_batch, _ in test_loader:\n",
333
- " output = model(x_num_batch.to(device), x_cat_batch.to(device))\n",
334
- " output = torch.softmax(output, dim=1)\n",
335
- " test_preds.extend(output.cpu().numpy())\n",
336
- "\n",
337
- "\n",
338
- " return test_preds\n"
339
- ]
340
- },
341
- {
342
- "cell_type": "code",
343
- "execution_count": 5,
344
- "metadata": {},
345
- "outputs": [],
346
- "source": [
347
- "df_seoul = pd.read_csv(\"../../data/data_for_modeling/seoul_train.csv\")\n",
348
- "df_seoul_test = pd.read_csv(\"../../data/data_for_modeling/seoul_test.csv\")\n",
349
- "\n",
350
- "df_busan = pd.read_csv(\"../../data/data_for_modeling/busan_train.csv\")\n",
351
- "df_busan_test = pd.read_csv(\"../../data/data_for_modeling/busan_test.csv\")\n",
352
- "\n",
353
- "df_daegu = pd.read_csv(\"../../data/data_for_modeling/daegu_train.csv\")\n",
354
- "df_daegu_test = pd.read_csv(\"../../data/data_for_modeling/daegu_test.csv\")\n",
355
- "\n",
356
- "df_daejeon = pd.read_csv(\"../../data/data_for_modeling/daejeon_train.csv\")\n",
357
- "df_daejeon_test = pd.read_csv(\"../../data/data_for_modeling/daejeon_test.csv\")\n",
358
- "\n",
359
- "df_incheon = pd.read_csv(\"../../data/data_for_modeling/incheon_train.csv\")\n",
360
- "df_incheon_test = pd.read_csv(\"../../data/data_for_modeling/incheon_test.csv\")\n",
361
- "\n",
362
- "df_gwangju = pd.read_csv(\"../../data/data_for_modeling/gwangju_train.csv\")\n",
363
- "df_gwangju_test = pd.read_csv(\"../../data/data_for_modeling/gwangju_test.csv\")"
364
- ]
365
- },
366
- {
367
- "cell_type": "code",
368
- "execution_count": 6,
369
- "metadata": {},
370
- "outputs": [],
371
- "source": [
372
- "def preprocessing_df(df):\n",
373
- " df = df[df.columns].copy()\n",
374
- " df['year'] = df['year'].astype('int')\n",
375
- " df['month'] = df['month'].astype('int')\n",
376
- " df['hour'] = df['hour'].astype('int')\n",
377
- " df['binary_class'] = df['binary_class'].astype('int')\n",
378
- " df['multi_class'] = df['multi_class'].astype('int')\n",
379
- "\n",
380
- " df.loc[df['wind_dir']=='정온', 'wind_dir'] = \"0\"\n",
381
- " df['wind_dir'] = df['wind_dir'].astype('int')\n",
382
- " df= df[['temp_C', 'precip_mm', 'wind_speed', 'wind_dir', 'hm',\n",
383
- " 'vap_pressure', 'dewpoint_C', 'loc_pressure', 'sea_pressure',\n",
384
- " 'solarRad', 'snow_cm', 'cloudcover', 'lm_cloudcover', 'low_cloudbase',\n",
385
- " 'groundtemp', 'O3', 'NO2', 'PM10', 'PM25', 'year',\n",
386
- " 'month', 'hour', 'ground_temp - temp_C', 'hour_sin', 'hour_cos',\n",
387
- " 'month_sin', 'month_cos','multi_class']]\n",
388
- " return df\n",
389
- "\n"
390
- ]
391
- },
392
- {
393
- "cell_type": "code",
394
- "execution_count": 7,
395
- "metadata": {},
396
- "outputs": [],
397
- "source": [
398
- "df_seoul_test= preprocessing_df(df_seoul_test).copy()\n",
399
- "df_busan_test= preprocessing_df(df_busan_test).copy()\n",
400
- "df_daegu_test= preprocessing_df(df_daegu_test).copy()\n",
401
- "df_gwangju_test= preprocessing_df(df_gwangju_test).copy()\n",
402
- "df_daejeon_test= preprocessing_df(df_daejeon_test).copy()\n",
403
- "df_incheon_test= preprocessing_df(df_incheon_test).copy()\n",
404
- "\n",
405
- "df_seoul= preprocessing_df(df_seoul).copy()\n",
406
- "df_busan= preprocessing_df(df_busan).copy()\n",
407
- "df_daegu= preprocessing_df(df_daegu).copy()\n",
408
- "df_gwangju= preprocessing_df(df_gwangju).copy()\n",
409
- "df_daejeon= preprocessing_df(df_daejeon).copy()\n",
410
- "df_incheon= preprocessing_df(df_incheon).copy()\n",
411
- "\n",
412
- "df_seoul_test.drop(columns=['year'], inplace=True)\n",
413
- "df_busan_test.drop(columns=['year'], inplace=True)\n",
414
- "df_daegu_test.drop(columns=['year'], inplace=True)\n",
415
- "df_daejeon_test.drop(columns=['year'], inplace=True)\n",
416
- "df_incheon_test.drop(columns=['year'], inplace=True)\n",
417
- "df_gwangju_test.drop(columns=['year'], inplace=True)"
418
- ]
419
- },
420
- {
421
- "cell_type": "code",
422
- "execution_count": 8,
423
- "metadata": {},
424
- "outputs": [],
425
- "source": [
426
- "import joblib\n",
427
- "\n",
428
- "lgb_seoul= joblib.load('../save_model/LGB_optima/lgb_seoul_smote.pkl')\n",
429
- "lgb_busan= joblib.load('../save_model/LGB_optima/lgb_busan_smote.pkl')\n",
430
- "lgb_incheon= joblib.load('../save_model/LGB_optima/lgb_incheon_smote.pkl')\n",
431
- "lgb_daegu= joblib.load('../save_model/LGB_optima/lgb_daegu_smote.pkl')\n",
432
- "lgb_daejeon= joblib.load('../save_model/LGB_optima/lgb_daejeon_smote.pkl')\n",
433
- "lgb_gwangju= joblib.load('../save_model/LGB_optima/lgb_gwangju_smote.pkl')\n",
434
- "\n",
435
- "xgb_seoul= joblib.load('../save_model/XGB_optima/xgb_seoul_smote.pkl')\n",
436
- "xgb_busan= joblib.load('../save_model/XGB_optima/xgb_busan_ctgan20000.pkl')\n",
437
- "xgb_incheon= joblib.load('../save_model/XGB_optima/xgb_incheon_smote.pkl')\n",
438
- "xgb_daegu= joblib.load('../save_model/XGB_optima/xgb_daegu_smote.pkl')\n",
439
- "xgb_daejeon= joblib.load('../save_model/XGB_optima/xgb_daejeon_smote.pkl')\n",
440
- "xgb_gwangju= joblib.load('../save_model/XGB_optima/xgb_gwangju_smote.pkl')\n",
441
- "\n",
442
- "lgb_seoul_1= lgb_seoul[0]\n",
443
- "lgb_seoul_2= lgb_seoul[1]\n",
444
- "lgb_seoul_3= lgb_seoul[2]\n",
445
- "\n",
446
- "lgb_busan_1= lgb_busan[0]\n",
447
- "lgb_busan_2= lgb_busan[1]\n",
448
- "lgb_busan_3= lgb_busan[2]\n",
449
- "\n",
450
- "lgb_incheon_1= lgb_incheon[0]\n",
451
- "lgb_incheon_2= lgb_incheon[1]\n",
452
- "lgb_incheon_3= lgb_incheon[2]\n",
453
- "\n",
454
- "lgb_daegu_1= lgb_daegu[0]\n",
455
- "lgb_daegu_2= lgb_daegu[1]\n",
456
- "lgb_daegu_3= lgb_daegu[2]\n",
457
- "\n",
458
- "lgb_daejeon_1= lgb_daejeon[0]\n",
459
- "lgb_daejeon_2= lgb_daejeon[1]\n",
460
- "lgb_daejeon_3= lgb_daejeon[2]\n",
461
- "\n",
462
- "lgb_gwangju_1= lgb_gwangju[0]\n",
463
- "lgb_gwangju_2= lgb_gwangju[1]\n",
464
- "lgb_gwangju_3= lgb_gwangju[2]\n",
465
- "\n",
466
- "\n",
467
- "xgb_seoul_1= xgb_seoul[0]\n",
468
- "xgb_seoul_2= xgb_seoul[1]\n",
469
- "xgb_seoul_3= xgb_seoul[2]\n",
470
- "\n",
471
- "xgb_busan_1= xgb_busan[0]\n",
472
- "xgb_busan_2= xgb_busan[1]\n",
473
- "xgb_busan_3= xgb_busan[2]\n",
474
- "\n",
475
- "xgb_incheon_1= xgb_incheon[0]\n",
476
- "xgb_incheon_2= xgb_incheon[1]\n",
477
- "xgb_incheon_3= xgb_incheon[2]\n",
478
- "\n",
479
- "xgb_daegu_1= xgb_daegu[0]\n",
480
- "xgb_daegu_2= xgb_daegu[1]\n",
481
- "xgb_daegu_3= xgb_daegu[2]\n",
482
- "\n",
483
- "xgb_daejeon_1= xgb_daejeon[0]\n",
484
- "xgb_daejeon_2= xgb_daejeon[1]\n",
485
- "xgb_daejeon_3= xgb_daejeon[2]\n",
486
- "\n",
487
- "xgb_gwangju_1= xgb_gwangju[0]\n",
488
- "xgb_gwangju_2= xgb_gwangju[1]\n",
489
- "xgb_gwangju_3= xgb_gwangju[2]\n"
490
- ]
491
- },
492
- {
493
- "cell_type": "markdown",
494
- "metadata": {},
495
- "source": [
496
- "## **Soft Voting**"
497
- ]
498
- },
499
- {
500
- "cell_type": "markdown",
501
- "metadata": {},
502
- "source": [
503
- "## **서울**"
504
- ]
505
- },
506
- {
507
- "cell_type": "code",
508
- "execution_count": 9,
509
- "metadata": {},
510
- "outputs": [],
511
- "source": [
512
- "voting = []\n",
513
- "mcc = []\n",
514
- "accuracy = []\n"
515
- ]
516
- },
517
- {
518
- "cell_type": "code",
519
- "execution_count": 10,
520
- "metadata": {},
521
- "outputs": [
522
- {
523
- "data": {
524
- "text/plain": [
525
- "Index(['temp_C', 'precip_mm', 'wind_speed', 'wind_dir', 'hm', 'vap_pressure',\n",
526
- " 'dewpoint_C', 'loc_pressure', 'sea_pressure', 'solarRad', 'snow_cm',\n",
527
- " 'cloudcover', 'lm_cloudcover', 'low_cloudbase', 'groundtemp', 'O3',\n",
528
- " 'NO2', 'PM10', 'PM25', 'month', 'hour', 'ground_temp - temp_C',\n",
529
- " 'hour_sin', 'hour_cos', 'month_sin', 'month_cos', 'multi_class'],\n",
530
- " dtype='object')"
531
- ]
532
- },
533
- "execution_count": 10,
534
- "metadata": {},
535
- "output_type": "execute_result"
536
- }
537
- ],
538
- "source": [
539
- "df_seoul_test.columns"
540
- ]
541
- },
542
- {
543
- "cell_type": "code",
544
- "execution_count": 11,
545
- "metadata": {},
546
- "outputs": [
547
- {
548
- "name": "stdout",
549
- "output_type": "stream",
550
- "text": [
551
- "CSI score of soft(test) : 0.3248062015503624\n"
552
- ]
553
- }
554
- ],
555
- "source": [
556
- "from scipy.stats import mode # 최다 득표수 계산을 위한 mode 함수\n",
557
- "\n",
558
- "probas = []\n",
559
- "\n",
560
- "# 1 Fold\n",
561
- "test_preds = get_proba('seoul', 'deepgbm', 'ctgan20000', 1)\n",
562
- "probas.append(test_preds)\n",
563
- "test_preds = get_proba('seoul', 'resnet_like', 'smote', 1)\n",
564
- "probas.append(test_preds)\n",
565
- "# probas.append(xgb_seoul_1.predict_proba(df_seoul_test.iloc[:,:-1]))\n",
566
- "\n",
567
- "# 2 Fold\n",
568
- "test_preds = get_proba('seoul', 'deepgbm', 'ctgan20000', 2)\n",
569
- "probas.append(test_preds)\n",
570
- "test_preds = get_proba('seoul', 'resnet_like', 'smote', 2)\n",
571
- "probas.append(test_preds)\n",
572
- "# probas.append(xgb_seoul_2.predict_proba(df_seoul_test.iloc[:,:-1]))\n",
573
- "\n",
574
- "# 3 Fold\n",
575
- "test_preds = get_proba('seoul', 'deepgbm', 'ctgan20000', 3)\n",
576
- "probas.append(test_preds)\n",
577
- "test_preds = get_proba('seoul', 'resnet_like', 'smote', 3)\n",
578
- "probas.append(test_preds)\n",
579
- "# probas.append(xgb_seoul_3.predict_proba(df_seoul_test.iloc[:,:-1]))\n",
580
- "\n",
581
- "voting.append(calculate_csi(df_seoul_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))\n",
582
- "mcc.append(multiclass_mcc(df_seoul_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))\n",
583
- "accuracy.append(accuracy_score(df_seoul_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))\n",
584
- "\n",
585
- "\n",
586
- "print(\"CSI score of soft(test) :\", calculate_csi(df_seoul_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))\n"
587
- ]
588
- },
589
- {
590
- "cell_type": "code",
591
- "execution_count": 12,
592
- "metadata": {},
593
- "outputs": [
594
- {
595
- "data": {
596
- "text/plain": [
597
- "array([[ 2, 11, 0],\n",
598
- " [ 6, 417, 58],\n",
599
- " [ 7, 789, 7470]])"
600
- ]
601
- },
602
- "execution_count": 12,
603
- "metadata": {},
604
- "output_type": "execute_result"
605
- }
606
- ],
607
- "source": [
608
- "confusion_matrix(df_seoul_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1))"
609
- ]
610
- },
611
- {
612
- "cell_type": "markdown",
613
- "metadata": {},
614
- "source": [
615
- "## **부산**"
616
- ]
617
- },
618
- {
619
- "cell_type": "code",
620
- "execution_count": 13,
621
- "metadata": {},
622
- "outputs": [
623
- {
624
- "name": "stdout",
625
- "output_type": "stream",
626
- "text": [
627
- "CSI score of soft(test) : 0.46608315098458075\n"
628
- ]
629
- }
630
- ],
631
- "source": [
632
- "from scipy.stats import mode # 최다 득표수 계산을 위한 mode 함수\n",
633
- "\n",
634
- "probas = []\n",
635
- "\n",
636
- "# 1 Fold\n",
637
- "test_preds = get_proba('busan', 'deepgbm', 'pure', 1)\n",
638
- "probas.append(test_preds)\n",
639
- "test_preds = get_proba('busan', 'resnet_like', 'ctgan10000', 1)\n",
640
- "probas.append(test_preds)\n",
641
- "\n",
642
- "\n",
643
- "# 2 Fold\n",
644
- "test_preds = get_proba('busan', 'deepgbm', 'pure', 2)\n",
645
- "probas.append(test_preds)\n",
646
- "test_preds = get_proba('busan', 'resnet_like', 'ctgan10000', 2)\n",
647
- "probas.append(test_preds)\n",
648
- "\n",
649
- "\n",
650
- "# 3 Fold\n",
651
- "test_preds = get_proba('busan', 'deepgbm', 'pure', 3)\n",
652
- "probas.append(test_preds)\n",
653
- "test_preds = get_proba('busan', 'resnet_like', 'ctgan10000', 3)\n",
654
- "probas.append(test_preds)\n",
655
- "\n",
656
- "\n",
657
- "voting.append(calculate_csi(df_busan_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))\n",
658
- "mcc.append(multiclass_mcc(df_busan_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))\n",
659
- "accuracy.append(accuracy_score(df_busan_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))\n",
660
- "\n",
661
- "print(\"CSI score of soft(test) :\", calculate_csi(df_busan_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))\n"
662
- ]
663
- },
664
- {
665
- "cell_type": "code",
666
- "execution_count": 14,
667
- "metadata": {},
668
- "outputs": [
669
- {
670
- "data": {
671
- "text/plain": [
672
- "array([[ 11, 13, 0],\n",
673
- " [ 11, 202, 68],\n",
674
- " [ 2, 150, 8303]])"
675
- ]
676
- },
677
- "execution_count": 14,
678
- "metadata": {},
679
- "output_type": "execute_result"
680
- }
681
- ],
682
- "source": [
683
- "confusion_matrix(df_busan_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1))"
684
- ]
685
- },
686
- {
687
- "cell_type": "markdown",
688
- "metadata": {},
689
- "source": [
690
- "## **인천**"
691
- ]
692
- },
693
- {
694
- "cell_type": "code",
695
- "execution_count": 15,
696
- "metadata": {},
697
- "outputs": [
698
- {
699
- "name": "stdout",
700
- "output_type": "stream",
701
- "text": [
702
- "CSI score of hard(test) : 0.572269457161506\n"
703
- ]
704
- }
705
- ],
706
- "source": [
707
- "from scipy.stats import mode # 최다 득표수 계산을 위한 mode 함수\n",
708
- "\n",
709
- "\n",
710
- "# 1 Fold\n",
711
- "probas = []\n",
712
- "test_preds = get_proba('incheon', 'deepgbm', 'pure', 1)\n",
713
- "probas.append(test_preds)\n",
714
- "test_preds = get_proba('incheon', 'resnet_like', 'smote', 1)\n",
715
- "probas.append(test_preds)\n",
716
- "test_preds = get_proba('incheon', 'ft_transformer', 'pure', 1)\n",
717
- "probas.append(test_preds)\n",
718
- "\n",
719
- "probas.append(lgb_incheon_1.predict_proba(df_incheon_test.iloc[:,:-1]))\n",
720
- "probas.append(xgb_incheon_1.predict_proba(df_incheon_test.iloc[:,:-1]))\n",
721
- "\n",
722
- "# 2 Fold\n",
723
- "test_preds = get_proba('incheon', 'deepgbm', 'pure', 2)\n",
724
- "probas.append(test_preds)\n",
725
- "test_preds = get_proba('incheon', 'resnet_like', 'smote', 2)\n",
726
- "probas.append(test_preds)\n",
727
- "test_preds = get_proba('incheon', 'ft_transformer', 'pure', 2)\n",
728
- "probas.append(test_preds)\n",
729
- "\n",
730
- "probas.append(lgb_incheon_2.predict_proba(df_incheon_test.iloc[:,:-1]))\n",
731
- "probas.append(xgb_incheon_2.predict_proba(df_incheon_test.iloc[:,:-1]))\n",
732
- "\n",
733
- "# 3 Fold\n",
734
- "test_preds = get_proba('incheon', 'deepgbm', 'pure', 3)\n",
735
- "probas.append(test_preds)\n",
736
- "test_preds = get_proba('incheon', 'resnet_like', 'smote', 3)\n",
737
- "probas.append(test_preds)\n",
738
- "test_preds = get_proba('incheon', 'ft_transformer', 'pure', 3)\n",
739
- "probas.append(test_preds)\n",
740
- "\n",
741
- "probas.append(lgb_incheon_3.predict_proba(df_incheon_test.iloc[:,:-1]))\n",
742
- "probas.append(xgb_incheon_3.predict_proba(df_incheon_test.iloc[:,:-1]))\n",
743
- "\n",
744
- "\n",
745
- "\n",
746
- "voting.append(calculate_csi(df_incheon_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))\n",
747
- "mcc.append(multiclass_mcc(df_incheon_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))\n",
748
- "accuracy.append(accuracy_score(df_incheon_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))\n",
749
- "\n",
750
- "\n",
751
- "print(\"CSI score of hard(test) :\", calculate_csi(df_incheon_test.iloc[:,-1],mode(np.argmax(probas, axis=2), axis=0).mode[0]))\n"
752
- ]
753
- },
754
- {
755
- "cell_type": "code",
756
- "execution_count": 16,
757
- "metadata": {},
758
- "outputs": [
759
- {
760
- "data": {
761
- "text/plain": [
762
- "array([[ 87, 74, 21],\n",
763
- " [ 22, 788, 395],\n",
764
- " [ 2, 140, 7231]])"
765
- ]
766
- },
767
- "execution_count": 16,
768
- "metadata": {},
769
- "output_type": "execute_result"
770
- }
771
- ],
772
- "source": [
773
- "confusion_matrix(df_incheon_test.iloc[:,-1],mode(np.argmax(probas, axis=2), axis=0).mode[0])"
774
- ]
775
- },
776
- {
777
- "cell_type": "markdown",
778
- "metadata": {},
779
- "source": [
780
- "## **대구**"
781
- ]
782
- },
783
- {
784
- "cell_type": "code",
785
- "execution_count": 17,
786
- "metadata": {},
787
- "outputs": [
788
- {
789
- "name": "stdout",
790
- "output_type": "stream",
791
- "text": [
792
- "CSI score of soft(test) : 0.2852112676055334\n"
793
- ]
794
- }
795
- ],
796
- "source": [
797
- "from scipy.stats import mode # 최다 득표수 계산을 위한 mode 함수\n",
798
- "\n",
799
- "probas= []\n",
800
- "\n",
801
- "# 1 Fold\n",
802
- "test_preds = get_proba('daegu', 'deepgbm', 'smote', 1)\n",
803
- "probas.append(test_preds)\n",
804
- "test_preds = get_proba('daegu', 'ft_transformer', 'pure', 1)\n",
805
- "probas.append(test_preds)\n",
806
- "\n",
807
- "# 2 Fold\n",
808
- "test_preds = get_proba('daegu', 'deepgbm', 'smote', 2)\n",
809
- "probas.append(test_preds)\n",
810
- "test_preds = get_proba('daegu', 'ft_transformer', 'pure', 2)\n",
811
- "probas.append(test_preds)\n",
812
- "\n",
813
- "# 3 Fold\n",
814
- "test_preds = get_proba('daegu', 'deepgbm', 'smote', 3)\n",
815
- "probas.append(test_preds)\n",
816
- "test_preds = get_proba('daegu', 'ft_transformer', 'pure', 3)\n",
817
- "probas.append(test_preds)\n",
818
- "\n",
819
- "voting.append(calculate_csi(df_daegu_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))\n",
820
- "mcc.append(multiclass_mcc(df_daegu_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))\n",
821
- "accuracy.append(accuracy_score(df_daegu_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))\n",
822
- "\n",
823
- "print(\"CSI score of soft(test) :\", calculate_csi(df_daegu_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))\n"
824
- ]
825
- },
826
- {
827
- "cell_type": "code",
828
- "execution_count": 18,
829
- "metadata": {},
830
- "outputs": [
831
- {
832
- "data": {
833
- "text/plain": [
834
- "array([[ 1, 0, 0],\n",
835
- " [ 1, 80, 47],\n",
836
- " [ 2, 153, 8476]])"
837
- ]
838
- },
839
- "execution_count": 18,
840
- "metadata": {},
841
- "output_type": "execute_result"
842
- }
843
- ],
844
- "source": [
845
- "confusion_matrix(df_daegu_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1))"
846
- ]
847
- },
848
- {
849
- "cell_type": "markdown",
850
- "metadata": {},
851
- "source": [
852
- "## **대전**"
853
- ]
854
- },
855
- {
856
- "cell_type": "code",
857
- "execution_count": 19,
858
- "metadata": {},
859
- "outputs": [
860
- {
861
- "name": "stdout",
862
- "output_type": "stream",
863
- "text": [
864
- "CSI score of soft(test) : 0.31884057971011603\n"
865
- ]
866
- }
867
- ],
868
- "source": [
869
- "from scipy.stats import mode # 최다 득표수 계산을 위한 mode 함수\n",
870
- "\n",
871
- "probas = []\n",
872
- "\n",
873
- "# 1 Fold\n",
874
- "test_preds = get_proba('daejeon', 'deepgbm', 'pure', 1)\n",
875
- "probas.append(test_preds)\n",
876
- "test_preds = get_proba('daejeon', 'ft_transformer', 'pure', 1)\n",
877
- "probas.append(test_preds)\n",
878
- "\n",
879
- "\n",
880
- "# 2 Fold\n",
881
- "test_preds = get_proba('daejeon', 'deepgbm', 'pure', 2)\n",
882
- "probas.append(test_preds)\n",
883
- "test_preds = get_proba('daejeon', 'ft_transformer', 'pure', 2)\n",
884
- "probas.append(test_preds)\n",
885
- "\n",
886
- "\n",
887
- "# 3 Fold\n",
888
- "test_preds = get_proba('daejeon', 'deepgbm', 'pure', 3)\n",
889
- "probas.append(test_preds)\n",
890
- "test_preds = get_proba('daejeon', 'ft_transformer', 'pure', 3)\n",
891
- "probas.append(test_preds)\n",
892
- "\n",
893
- "voting.append(calculate_csi(df_daejeon_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))\n",
894
- "mcc.append(multiclass_mcc(df_daejeon_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))\n",
895
- "accuracy.append(accuracy_score(df_daejeon_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))\n",
896
- "\n",
897
- "\n",
898
- "print(\"CSI score of soft(test) :\", calculate_csi(df_daejeon_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))"
899
- ]
900
- },
901
- {
902
- "cell_type": "code",
903
- "execution_count": 20,
904
- "metadata": {},
905
- "outputs": [
906
- {
907
- "data": {
908
- "text/plain": [
909
- "array([[ 15, 23, 15],\n",
910
- " [ 10, 337, 271],\n",
911
- " [ 0, 433, 7656]])"
912
- ]
913
- },
914
- "execution_count": 20,
915
- "metadata": {},
916
- "output_type": "execute_result"
917
- }
918
- ],
919
- "source": [
920
- "confusion_matrix(df_daejeon_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1))"
921
- ]
922
- },
923
- {
924
- "cell_type": "markdown",
925
- "metadata": {},
926
- "source": [
927
- "## **광주**"
928
- ]
929
- },
930
- {
931
- "cell_type": "code",
932
- "execution_count": 21,
933
- "metadata": {},
934
- "outputs": [
935
- {
936
- "name": "stdout",
937
- "output_type": "stream",
938
- "text": [
939
- "CSI score of soft(test) : 0.4759725400457121\n"
940
- ]
941
- }
942
- ],
943
- "source": [
944
- "from scipy.stats import mode # 최다 득표수 계산을 위한 mode 함수\n",
945
- "\n",
946
- "probas = []\n",
947
- "\n",
948
- "# 1 Fold\n",
949
- "test_preds = get_proba('gwangju', 'deepgbm', 'pure', 1)\n",
950
- "probas.append(test_preds)\n",
951
- "test_preds = get_proba('gwangju', 'ft_transformer', 'pure', 1)\n",
952
- "probas.append(test_preds)\n",
953
- "\n",
954
- "\n",
955
- "# 2 Fold\n",
956
- "test_preds = get_proba('gwangju', 'deepgbm', 'pure', 2)\n",
957
- "probas.append(test_preds)\n",
958
- "test_preds = get_proba('gwangju', 'ft_transformer', 'pure', 2)\n",
959
- "probas.append(test_preds)\n",
960
- "\n",
961
- "\n",
962
- "# 3 Fold\n",
963
- "test_preds = get_proba('gwangju', 'deepgbm', 'pure', 3)\n",
964
- "probas.append(test_preds)\n",
965
- "test_preds = get_proba('gwangju', 'ft_transformer', 'pure', 3)\n",
966
- "probas.append(test_preds)\n",
967
- "\n",
968
- "voting.append(calculate_csi(df_gwangju_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))\n",
969
- "mcc.append(multiclass_mcc(df_gwangju_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))\n",
970
- "accuracy.append(accuracy_score(df_gwangju_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))\n",
971
- "\n",
972
- "\n",
973
- "print(\"CSI score of soft(test) :\", calculate_csi(df_gwangju_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1)))"
974
- ]
975
- },
976
- {
977
- "cell_type": "code",
978
- "execution_count": 22,
979
- "metadata": {},
980
- "outputs": [
981
- {
982
- "data": {
983
- "text/plain": [
984
- "array([[ 10, 12, 8],\n",
985
- " [ 2, 406, 235],\n",
986
- " [ 0, 201, 7886]])"
987
- ]
988
- },
989
- "execution_count": 22,
990
- "metadata": {},
991
- "output_type": "execute_result"
992
- }
993
- ],
994
- "source": [
995
- "confusion_matrix(df_gwangju_test.iloc[:,-1], np.argmax(np.mean(probas, axis=0), axis=1))"
996
- ]
997
- },
998
- {
999
- "cell_type": "code",
1000
- "execution_count": 23,
1001
- "metadata": {},
1002
- "outputs": [
1003
- {
1004
- "data": {
1005
- "image/png": "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",
1006
- "text/plain": [
1007
- "<Figure size 1000x500 with 1 Axes>"
1008
- ]
1009
- },
1010
- "metadata": {},
1011
- "output_type": "display_data"
1012
- }
1013
- ],
1014
- "source": [
1015
- "import matplotlib.pyplot as plt\n",
1016
- "import seaborn as sns\n",
1017
- "\n",
1018
- "plt.figure(figsize=(10,5))\n",
1019
- "sns.barplot(x=['Seoul', 'Busan', 'Incheon', 'Daegu', 'Daejeon', 'Gwangju'], y=voting)\n",
1020
- "plt.title('CSI Score of Voting Model')\n",
1021
- "plt.ylabel('CSI')\n",
1022
- "\n",
1023
- "plt.show()"
1024
- ]
1025
- },
1026
- {
1027
- "cell_type": "code",
1028
- "execution_count": 24,
1029
- "metadata": {},
1030
- "outputs": [
1031
- {
1032
- "data": {
1033
- "text/plain": [
1034
- "0.4078715882283252"
1035
- ]
1036
- },
1037
- "execution_count": 24,
1038
- "metadata": {},
1039
- "output_type": "execute_result"
1040
- }
1041
- ],
1042
- "source": [
1043
- "np.mean(voting)"
1044
- ]
1045
- },
1046
- {
1047
- "cell_type": "code",
1048
- "execution_count": 25,
1049
- "metadata": {},
1050
- "outputs": [
1051
- {
1052
- "data": {
1053
- "text/plain": [
1054
- "[0.3248062015503624,\n",
1055
- " 0.46608315098458075,\n",
1056
- " 0.5763157894736463,\n",
1057
- " 0.2852112676055334,\n",
1058
- " 0.31884057971011603,\n",
1059
- " 0.4759725400457121]"
1060
- ]
1061
- },
1062
- "execution_count": 25,
1063
- "metadata": {},
1064
- "output_type": "execute_result"
1065
- }
1066
- ],
1067
- "source": [
1068
- "voting"
1069
- ]
1070
- },
1071
- {
1072
- "cell_type": "code",
1073
- "execution_count": 26,
1074
- "metadata": {},
1075
- "outputs": [
1076
- {
1077
- "data": {
1078
- "text/plain": [
1079
- "[0.5106142349456536,\n",
1080
- " 0.640202275543952,\n",
1081
- " 0.709448778435959,\n",
1082
- " 0.45579515959653394,\n",
1083
- " 0.453960121993875,\n",
1084
- " 0.6218724605270242]"
1085
- ]
1086
- },
1087
- "execution_count": 26,
1088
- "metadata": {},
1089
- "output_type": "execute_result"
1090
- }
1091
- ],
1092
- "source": [
1093
- "mcc"
1094
- ]
1095
- },
1096
- {
1097
- "cell_type": "code",
1098
- "execution_count": 27,
1099
- "metadata": {},
1100
- "outputs": [
1101
- {
1102
- "data": {
1103
- "text/plain": [
1104
- "[0.9005707762557078,\n",
1105
- " 0.9721461187214612,\n",
1106
- " 0.9264840182648402,\n",
1107
- " 0.9768264840182649,\n",
1108
- " 0.9141552511415525,\n",
1109
- " 0.9477168949771689]"
1110
- ]
1111
- },
1112
- "execution_count": 27,
1113
- "metadata": {},
1114
- "output_type": "execute_result"
1115
- }
1116
- ],
1117
- "source": [
1118
- "accuracy"
1119
- ]
1120
- }
1121
- ],
1122
- "metadata": {
1123
- "kernelspec": {
1124
- "display_name": "Python 3",
1125
- "language": "python",
1126
- "name": "python3"
1127
- },
1128
- "language_info": {
1129
- "codemirror_mode": {
1130
- "name": "ipython",
1131
- "version": 3
1132
- },
1133
- "file_extension": ".py",
1134
- "mimetype": "text/x-python",
1135
- "name": "python",
1136
- "nbconvert_exporter": "python",
1137
- "pygments_lexer": "ipython3",
1138
- "version": "3.8.10"
1139
- }
1140
- },
1141
- "nbformat": 4,
1142
- "nbformat_minor": 2
1143
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/find_reason/ busan_trend.ipynb DELETED
The diff for this file is too large to render. See raw diff
 
Analysis_code/find_reason/ daegu_trend.ipynb DELETED
The diff for this file is too large to render. See raw diff
 
Analysis_code/find_reason/ gwangju_trend.ipynb DELETED
The diff for this file is too large to render. See raw diff
 
Analysis_code/find_reason/ incheon_trend.ipynb DELETED
The diff for this file is too large to render. See raw diff
 
Analysis_code/find_reason/ seoul_trend.ipynb DELETED
The diff for this file is too large to render. See raw diff
 
Analysis_code/find_reason/daejeon_trend.ipynb DELETED
The diff for this file is too large to render. See raw diff
 
Analysis_code/find_reason/make_trend_plot.ipynb DELETED
The diff for this file is too large to render. See raw diff
 
Analysis_code/find_reason/wasserstein_distance.ipynb DELETED
@@ -1,541 +0,0 @@
1
- {
2
- "cells": [
3
- {
4
- "cell_type": "code",
5
- "execution_count": 6,
6
- "metadata": {},
7
- "outputs": [],
8
- "source": [
9
- "# 분석에 필요한 라이브러리 임포트\n",
10
- "import warnings\n",
11
- "warnings.filterwarnings('ignore')\n",
12
- "import pandas as pd\n",
13
- "import numpy as np\n",
14
- "import matplotlib.pyplot as plt\n",
15
- "import seaborn as sns\n",
16
- "from scipy import stats\n",
17
- "from scipy.spatial import distance\n",
18
- "from scipy.stats import wasserstein_distance, entropy, ks_2samp\n",
19
- "from sklearn.manifold import TSNE\n",
20
- "from sklearn.preprocessing import StandardScaler\n",
21
- "from sklearn.ensemble import RandomForestRegressor\n",
22
- "from sklearn.ensemble import RandomForestClassifier # Added\n",
23
- "from sklearn.model_selection import train_test_split # Added\n",
24
- "from sklearn.metrics import roc_auc_score # Added\n",
25
- "from statsmodels.distributions.empirical_distribution import ECDF # Added\n",
26
- "import ot\n",
27
- "\n",
28
- "\n",
29
- "# 한글 폰트 설정\n",
30
- "plt.rcParams['font.family'] = 'NanumGothic'\n",
31
- "plt.rcParams['axes.unicode_minus'] = False"
32
- ]
33
- },
34
- {
35
- "cell_type": "code",
36
- "execution_count": 7,
37
- "metadata": {},
38
- "outputs": [],
39
- "source": [
40
- "seoul = pd.read_feather(\"../../data/data_for_modeling/df_seoul.feather\")\n",
41
- "seoul= seoul[['datetime','hm','PM10','PM25','year','month','hour','multi_class']]\n",
42
- "\n",
43
- "busan = pd.read_feather(\"../../data/data_for_modeling/df_busan.feather\")\n",
44
- "busan= busan[['datetime','hm','PM10','PM25','year','month','hour','multi_class']]\n",
45
- "\n",
46
- "incheon = pd.read_feather(\"../../data/data_for_modeling/df_incheon.feather\")\n",
47
- "incheon= incheon[['datetime','hm','PM10','PM25','year','month','hour','multi_class']]\n",
48
- "\n",
49
- "daegu = pd.read_feather(\"../../data/data_for_modeling/df_daegu.feather\")\n",
50
- "daegu= daegu[['datetime','hm','PM10','PM25','year','month','hour','multi_class']]\n",
51
- "\n",
52
- "daejeon = pd.read_feather(\"../../data/data_for_modeling/df_daejeon.feather\")\n",
53
- "daejeon= daejeon[['datetime','hm','PM10','PM25','year','month','hour','multi_class']]\n",
54
- "\n",
55
- "gwangju = pd.read_feather(\"../../data/data_for_modeling/df_gwangju.feather\")\n",
56
- "gwangju= gwangju[['datetime','hm','PM10','PM25','year','month','hour','multi_class']]"
57
- ]
58
- },
59
- {
60
- "cell_type": "code",
61
- "execution_count": 8,
62
- "metadata": {},
63
- "outputs": [
64
- {
65
- "name": "stdout",
66
- "output_type": "stream",
67
- "text": [
68
- "[0.5920662 0.92351786]\n",
69
- "[0.60414398 0.9190468 ]\n",
70
- "[0.60250035 0.9391276 ]\n",
71
- "[0.60112832 0.92493121]\n",
72
- "[0.58469137 0.90476229]\n",
73
- "[0.617718 0.93503164]\n"
74
- ]
75
- }
76
- ],
77
- "source": [
78
- "from sklearn.decomposition import PCA\n",
79
- "\n",
80
- "# 특성 선택 (예: PM10, PM25, hm 등)\n",
81
- "features = ['PM10','PM25', 'hm']\n",
82
- "# 스케일링\n",
83
- "scaler = StandardScaler()\n",
84
- "scaled_features = scaler.fit_transform(seoul[features])\n",
85
- "pca = PCA(n_components=2)\n",
86
- "pca.fit(scaled_features)\n",
87
- "print(pca.explained_variance_ratio_.cumsum())\n",
88
- "seoul_pca = pca.transform(scaled_features)\n",
89
- "seoul.drop(columns=['PM25', 'hm'], inplace=True)\n",
90
- "seoul[['pca_x', 'pca_y']] = seoul_pca\n",
91
- "\n",
92
- "\n",
93
- "scaled_features = scaler.fit_transform(busan[features])\n",
94
- "pca = PCA(n_components=2)\n",
95
- "pca.fit(scaled_features)\n",
96
- "print(pca.explained_variance_ratio_.cumsum())\n",
97
- "busan_pca = pca.transform(scaled_features)\n",
98
- "busan.drop(columns=['PM25', 'hm'], inplace=True)\n",
99
- "busan[['pca_x', 'pca_y']] = busan_pca\n",
100
- "\n",
101
- "scaled_features = scaler.fit_transform(incheon[features]) \n",
102
- "pca = PCA(n_components=2)\n",
103
- "pca.fit(scaled_features)\n",
104
- "print(pca.explained_variance_ratio_.cumsum())\n",
105
- "incheon_pca = pca.transform(scaled_features)\n",
106
- "incheon.drop(columns=['PM25', 'hm'], inplace=True)\n",
107
- "incheon[['pca_x', 'pca_y']] = incheon_pca\n",
108
- "\n",
109
- "scaled_features = scaler.fit_transform(daegu[features])\n",
110
- "pca = PCA(n_components=2)\n",
111
- "pca.fit(scaled_features)\n",
112
- "print(pca.explained_variance_ratio_.cumsum())\n",
113
- "daegu_pca = pca.transform(scaled_features)\n",
114
- "daegu.drop(columns=['PM25', 'hm'], inplace=True)\n",
115
- "daegu[['pca_x', 'pca_y']] = daegu_pca\n",
116
- "\n",
117
- "scaled_features = scaler.fit_transform(daejeon[features])\n",
118
- "pca = PCA(n_components=2)\n",
119
- "pca.fit(scaled_features)\n",
120
- "print(pca.explained_variance_ratio_.cumsum())\n",
121
- "daejeon_pca = pca.transform(scaled_features)\n",
122
- "daejeon.drop(columns=['PM25', 'hm'], inplace=True)\n",
123
- "daejeon[['pca_x', 'pca_y']] = daejeon_pca\n",
124
- "\n",
125
- "scaled_features = scaler.fit_transform(gwangju[features])\n",
126
- "pca = PCA(n_components=2)\n",
127
- "pca.fit(scaled_features)\n",
128
- "print(pca.explained_variance_ratio_.cumsum())\n",
129
- "gwangju_pca = pca.transform(scaled_features)\n",
130
- "gwangju.drop(columns=['PM25', 'hm'], inplace=True)\n",
131
- "gwangju[['pca_x', 'pca_y']] = gwangju_pca\n"
132
- ]
133
- },
134
- {
135
- "cell_type": "code",
136
- "execution_count": 31,
137
- "metadata": {},
138
- "outputs": [],
139
- "source": [
140
- "seoul_2018 = seoul[seoul['year'] == 2018]\n",
141
- "seoul_2019 = seoul[seoul['year'] == 2019]\n",
142
- "seoul_2020 = seoul[seoul['year'] == 2020]\n",
143
- "seoul_2021 = seoul[seoul['year'] == 2021]\n",
144
- "years = [2018, 2019, 2020, 2021]\n",
145
- "\n",
146
- "\n",
147
- "busan_2018 = busan[busan['year'] == 2018]\n",
148
- "busan_2019 = busan[busan['year'] == 2019]\n",
149
- "busan_2020 = busan[busan['year'] == 2020]\n",
150
- "busan_2021 = busan[busan['year'] == 2021]\n",
151
- "years = [2018, 2019, 2020, 2021]\n",
152
- "\n",
153
- "\n",
154
- "incheon_2018 = incheon[incheon['year'] == 2018]\n",
155
- "incheon_2019 = incheon[incheon['year'] == 2019]\n",
156
- "incheon_2020 = incheon[incheon['year'] == 2020]\n",
157
- "incheon_2021 = incheon[incheon['year'] == 2021]\n",
158
- "years = [2018, 2019, 2020, 2021]\n",
159
- "\n",
160
- "\n",
161
- "daegu_2018 = daegu[daegu['year'] == 2018]\n",
162
- "daegu_2019 = daegu[daegu['year'] == 2019]\n",
163
- "daegu_2020 = daegu[daegu['year'] == 2020]\n",
164
- "daegu_2021 = daegu[daegu['year'] == 2021]\n",
165
- "years = [2018, 2019, 2020, 2021]\n",
166
- "\n",
167
- "\n",
168
- "daejeon_2018 = daejeon[daejeon['year'] == 2018]\n",
169
- "daejeon_2019 = daejeon[daejeon['year'] == 2019]\n",
170
- "daejeon_2020 = daejeon[daejeon['year'] == 2020]\n",
171
- "daejeon_2021 = daejeon[daejeon['year'] == 2021]\n",
172
- "years = [2018, 2019, 2020, 2021]\n",
173
- "\n",
174
- "\n",
175
- "gwangju_2018 = gwangju[gwangju['year'] == 2018]\n",
176
- "gwangju_2019 = gwangju[gwangju['year'] == 2019]\n",
177
- "gwangju_2020 = gwangju[gwangju['year'] == 2020]\n",
178
- "gwangju_2021 = gwangju[gwangju['year'] == 2021]\n",
179
- "years = [2018, 2019, 2020, 2021]\n",
180
- "\n",
181
- "\n",
182
- "\n"
183
- ]
184
- },
185
- {
186
- "cell_type": "code",
187
- "execution_count": 33,
188
- "metadata": {},
189
- "outputs": [
190
- {
191
- "name": "stdout",
192
- "output_type": "stream",
193
- "text": [
194
- " 2018 2019 2020 2021\n",
195
- "2018 0.0 0.130217 0.063132 1.081307\n",
196
- "2019 0.130217 0.0 0.059051 0.830648\n",
197
- "2020 0.063132 0.059051 0.0 0.039927\n",
198
- "2021 1.081307 0.830648 0.039927 0.0\n"
199
- ]
200
- }
201
- ],
202
- "source": [
203
- "# 연도별 데이터 준비\n",
204
- "years = [2018, 2019, 2020, 2021]\n",
205
- "data_dict = {\n",
206
- " 2018: seoul_2018[['pca_x', 'pca_y']].values,\n",
207
- " 2019: seoul_2019[['pca_x', 'pca_y']].values,\n",
208
- " 2020: seoul_2020[['pca_x', 'pca_y']].values,\n",
209
- " 2021: seoul_2021[['pca_x', 'pca_y']].values\n",
210
- "}\n",
211
- "\n",
212
- "\n",
213
- "# 결과를 저장할 데이터프레임 생성\n",
214
- "result_df = pd.DataFrame(index=years, columns=years)\n",
215
- "\n",
216
- "for i, year1 in enumerate(years):\n",
217
- " for j, year2 in enumerate(years):\n",
218
- " if year1 == year2:\n",
219
- " result_df.iloc[i, j] = 0.0\n",
220
- " if j < i:\n",
221
- " # 이미 계산된 값 사용\n",
222
- " result_df.iloc[i, j] = result_df.iloc[j, i]\n",
223
- " else:\n",
224
- " X = data_dict[year1]\n",
225
- " Y = data_dict[year2]\n",
226
- " a = np.ones(len(X)) / len(X)\n",
227
- " b = np.ones(len(Y)) / len(Y)\n",
228
- " W = ot.emd2(a, b, ot.dist(X, Y))\n",
229
- " result_df.iloc[i, j] = W\n",
230
- " result_df.iloc[j, i] = W # 대칭 위치에 동일 값 저장\n",
231
- "\n",
232
- "# 결과 출력\n",
233
- "print(result_df)"
234
- ]
235
- },
236
- {
237
- "cell_type": "code",
238
- "execution_count": 23,
239
- "metadata": {},
240
- "outputs": [
241
- {
242
- "name": "stdout",
243
- "output_type": "stream",
244
- "text": [
245
- " 2018 2019 2020 2021\n",
246
- "2018 0.0 0.116261 0.10445 1.424479\n",
247
- "2019 0.116261 0.0 0.09933 1.164067\n",
248
- "2020 0.10445 0.09933 0.0 1.075336\n",
249
- "2021 1.424479 1.164067 1.075336 0.0\n"
250
- ]
251
- }
252
- ],
253
- "source": [
254
- "# 연도별 데이터 준비\n",
255
- "years = [2018, 2019, 2020, 2021]\n",
256
- "data_dict = {\n",
257
- " 2018: busan_2018[['pca_x', 'pca_y']].values,\n",
258
- " 2019: busan_2019[['pca_x', 'pca_y']].values,\n",
259
- " 2020: busan_2020[['pca_x', 'pca_y']].values,\n",
260
- " 2021: busan_2021[['pca_x', 'pca_y']].values\n",
261
- "}\n",
262
- "\n",
263
- "\n",
264
- "# 결과를 저장할 데이터프레임 생성\n",
265
- "result_df = pd.DataFrame(index=years, columns=years)\n",
266
- "\n",
267
- "for i, year1 in enumerate(years):\n",
268
- " for j, year2 in enumerate(years):\n",
269
- " if year1 == year2:\n",
270
- " result_df.iloc[i, j] = 0.0\n",
271
- " if j < i:\n",
272
- " # 이미 계산된 값 사용\n",
273
- " result_df.iloc[i, j] = result_df.iloc[j, i]\n",
274
- " else:\n",
275
- " X = data_dict[year1]\n",
276
- " Y = data_dict[year2]\n",
277
- " a = np.ones(len(X)) / len(X)\n",
278
- " b = np.ones(len(Y)) / len(Y)\n",
279
- " W = ot.emd2(a, b, ot.dist(X, Y))\n",
280
- " result_df.iloc[i, j] = W\n",
281
- " result_df.iloc[j, i] = W # 대칭 위치에 동일 값 저장\n",
282
- "\n",
283
- "# 결과 출력\n",
284
- "print(result_df)"
285
- ]
286
- },
287
- {
288
- "cell_type": "code",
289
- "execution_count": 24,
290
- "metadata": {},
291
- "outputs": [
292
- {
293
- "name": "stdout",
294
- "output_type": "stream",
295
- "text": [
296
- " 2018 2019 2020 2021\n",
297
- "2018 0.0 0.080291 0.074071 0.449094\n",
298
- "2019 0.080291 0.0 0.060171 0.384189\n",
299
- "2020 0.074071 0.060171 0.0 0.04047\n",
300
- "2021 0.449094 0.384189 0.04047 0.0\n"
301
- ]
302
- }
303
- ],
304
- "source": [
305
- "# 연도별 데이터 준비\n",
306
- "years = [2018, 2019, 2020, 2021]\n",
307
- "data_dict = {\n",
308
- " 2018: incheon_2018[['pca_x', 'pca_y']].values,\n",
309
- " 2019: incheon_2019[['pca_x', 'pca_y']].values,\n",
310
- " 2020: incheon_2020[['pca_x', 'pca_y']].values,\n",
311
- " 2021: incheon_2021[['pca_x', 'pca_y']].values\n",
312
- "}\n",
313
- "\n",
314
- "\n",
315
- "# 결과를 저장할 데이터프레임 생성\n",
316
- "result_df = pd.DataFrame(index=years, columns=years)\n",
317
- "\n",
318
- "for i, year1 in enumerate(years):\n",
319
- " for j, year2 in enumerate(years):\n",
320
- " if year1 == year2:\n",
321
- " result_df.iloc[i, j] = 0.0\n",
322
- " if j < i:\n",
323
- " # 이미 계산된 값 사용\n",
324
- " result_df.iloc[i, j] = result_df.iloc[j, i]\n",
325
- " else:\n",
326
- " X = data_dict[year1]\n",
327
- " Y = data_dict[year2]\n",
328
- " a = np.ones(len(X)) / len(X)\n",
329
- " b = np.ones(len(Y)) / len(Y)\n",
330
- " W = ot.emd2(a, b, ot.dist(X, Y))\n",
331
- " result_df.iloc[i, j] = W\n",
332
- " result_df.iloc[j, i] = W # 대칭 위치에 동일 값 저장\n",
333
- "\n",
334
- "# 결과 출력\n",
335
- "print(result_df)"
336
- ]
337
- },
338
- {
339
- "cell_type": "code",
340
- "execution_count": 25,
341
- "metadata": {},
342
- "outputs": [
343
- {
344
- "name": "stdout",
345
- "output_type": "stream",
346
- "text": [
347
- " 2018 2019 2020 2021\n",
348
- "2018 0.0 0.127512 0.112157 0.731476\n",
349
- "2019 0.127512 0.0 0.094651 0.647071\n",
350
- "2020 0.112157 0.094651 0.0 0.041217\n",
351
- "2021 0.731476 0.647071 0.041217 0.0\n"
352
- ]
353
- }
354
- ],
355
- "source": [
356
- "# 연도별 데이터 준비\n",
357
- "years = [2018, 2019, 2020, 2021]\n",
358
- "data_dict = {\n",
359
- " 2018: daegu_2018[['pca_x', 'pca_y']].values,\n",
360
- " 2019: daegu_2019[['pca_x', 'pca_y']].values,\n",
361
- " 2020: daegu_2020[['pca_x', 'pca_y']].values,\n",
362
- " 2021: daegu_2021[['pca_x', 'pca_y']].values\n",
363
- "}\n",
364
- "\n",
365
- "\n",
366
- "# 결과를 저장할 데이터프레임 생성\n",
367
- "result_df = pd.DataFrame(index=years, columns=years)\n",
368
- "\n",
369
- "for i, year1 in enumerate(years):\n",
370
- " for j, year2 in enumerate(years):\n",
371
- " if year1 == year2:\n",
372
- " result_df.iloc[i, j] = 0.0\n",
373
- " if j < i:\n",
374
- " # 이미 계산된 값 사용\n",
375
- " result_df.iloc[i, j] = result_df.iloc[j, i]\n",
376
- " else:\n",
377
- " X = data_dict[year1]\n",
378
- " Y = data_dict[year2]\n",
379
- " a = np.ones(len(X)) / len(X)\n",
380
- " b = np.ones(len(Y)) / len(Y)\n",
381
- " W = ot.emd2(a, b, ot.dist(X, Y))\n",
382
- " result_df.iloc[i, j] = W\n",
383
- " result_df.iloc[j, i] = W # 대칭 위치에 동일 값 저장\n",
384
- "\n",
385
- "# 결과 출력\n",
386
- "print(result_df)"
387
- ]
388
- },
389
- {
390
- "cell_type": "code",
391
- "execution_count": 26,
392
- "metadata": {},
393
- "outputs": [
394
- {
395
- "name": "stdout",
396
- "output_type": "stream",
397
- "text": [
398
- " 2018 2019 2020 2021\n",
399
- "2018 0.0 0.273013 0.053969 0.877338\n",
400
- "2019 0.273013 0.0 0.137817 0.780071\n",
401
- "2020 0.053969 0.137817 0.0 0.042294\n",
402
- "2021 0.877338 0.780071 0.042294 0.0\n"
403
- ]
404
- }
405
- ],
406
- "source": [
407
- "# 연도별 데이터 준비\n",
408
- "years = [2018, 2019, 2020, 2021]\n",
409
- "data_dict = {\n",
410
- " 2018: daejeon_2018[['pca_x', 'pca_y']].values,\n",
411
- " 2019: daejeon_2019[['pca_x', 'pca_y']].values,\n",
412
- " 2020: daejeon_2020[['pca_x', 'pca_y']].values,\n",
413
- " 2021: daejeon_2021[['pca_x', 'pca_y']].values\n",
414
- "}\n",
415
- "\n",
416
- "\n",
417
- "# 결과를 저장할 데이터프레임 생성\n",
418
- "result_df = pd.DataFrame(index=years, columns=years)\n",
419
- "\n",
420
- "for i, year1 in enumerate(years):\n",
421
- " for j, year2 in enumerate(years):\n",
422
- " if year1 == year2:\n",
423
- " result_df.iloc[i, j] = 0.0\n",
424
- " if j < i:\n",
425
- " # 이미 계산된 값 사용\n",
426
- " result_df.iloc[i, j] = result_df.iloc[j, i]\n",
427
- " else:\n",
428
- " X = data_dict[year1]\n",
429
- " Y = data_dict[year2]\n",
430
- " a = np.ones(len(X)) / len(X)\n",
431
- " b = np.ones(len(Y)) / len(Y)\n",
432
- " W = ot.emd2(a, b, ot.dist(X, Y))\n",
433
- " result_df.iloc[i, j] = W\n",
434
- " result_df.iloc[j, i] = W # 대칭 위치에 동일 값 저장\n",
435
- "\n",
436
- "# 결과 출력\n",
437
- "print(result_df)"
438
- ]
439
- },
440
- {
441
- "cell_type": "code",
442
- "execution_count": 27,
443
- "metadata": {},
444
- "outputs": [
445
- {
446
- "name": "stdout",
447
- "output_type": "stream",
448
- "text": [
449
- " 2018 2019 2020 2021\n",
450
- "2018 0.0 0.105633 0.08202 1.00155\n",
451
- "2019 0.105633 0.0 0.069322 0.892938\n",
452
- "2020 0.08202 0.069322 0.0 0.480667\n",
453
- "2021 1.00155 0.892938 0.480667 0.0\n"
454
- ]
455
- }
456
- ],
457
- "source": [
458
- "# 연도별 데이터 준비\n",
459
- "years = [2018, 2019, 2020, 2021]\n",
460
- "data_dict = {\n",
461
- " 2018: gwangju_2018[['pca_x', 'pca_y']].values,\n",
462
- " 2019: gwangju_2019[['pca_x', 'pca_y']].values,\n",
463
- " 2020: gwangju_2020[['pca_x', 'pca_y']].values,\n",
464
- " 2021: gwangju_2021[['pca_x', 'pca_y']].values\n",
465
- "}\n",
466
- "\n",
467
- "\n",
468
- "# 결과를 저장할 데이터프레임 생성\n",
469
- "result_df = pd.DataFrame(index=years, columns=years)\n",
470
- "\n",
471
- "for i, year1 in enumerate(years):\n",
472
- " for j, year2 in enumerate(years):\n",
473
- " if year1 == year2:\n",
474
- " result_df.iloc[i, j] = 0.0\n",
475
- " if j < i:\n",
476
- " # 이미 계산된 값 사용\n",
477
- " result_df.iloc[i, j] = result_df.iloc[j, i]\n",
478
- " else:\n",
479
- " X = data_dict[year1]\n",
480
- " Y = data_dict[year2]\n",
481
- " a = np.ones(len(X)) / len(X)\n",
482
- " b = np.ones(len(Y)) / len(Y)\n",
483
- " W = ot.emd2(a, b, ot.dist(X, Y))\n",
484
- " result_df.iloc[i, j] = W\n",
485
- " result_df.iloc[j, i] = W # 대칭 위치에 동일 값 저장\n",
486
- "\n",
487
- "# 결과 출력\n",
488
- "print(result_df)"
489
- ]
490
- },
491
- {
492
- "cell_type": "code",
493
- "execution_count": null,
494
- "metadata": {},
495
- "outputs": [],
496
- "source": []
497
- },
498
- {
499
- "cell_type": "code",
500
- "execution_count": null,
501
- "metadata": {},
502
- "outputs": [],
503
- "source": []
504
- },
505
- {
506
- "cell_type": "code",
507
- "execution_count": null,
508
- "metadata": {},
509
- "outputs": [],
510
- "source": []
511
- },
512
- {
513
- "cell_type": "code",
514
- "execution_count": null,
515
- "metadata": {},
516
- "outputs": [],
517
- "source": []
518
- }
519
- ],
520
- "metadata": {
521
- "kernelspec": {
522
- "display_name": "py39",
523
- "language": "python",
524
- "name": "python3"
525
- },
526
- "language_info": {
527
- "codemirror_mode": {
528
- "name": "ipython",
529
- "version": 3
530
- },
531
- "file_extension": ".py",
532
- "mimetype": "text/x-python",
533
- "name": "python",
534
- "nbconvert_exporter": "python",
535
- "pygments_lexer": "ipython3",
536
- "version": "3.9.18"
537
- }
538
- },
539
- "nbformat": 4,
540
- "nbformat_minor": 4
541
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/ft_transformer.py DELETED
@@ -1,56 +0,0 @@
1
- import torch
2
- import torch.nn as nn
3
-
4
- # FT-Transformer Implementation
5
- class FTTransformer(nn.Module):
6
- def __init__(self, num_features, cat_cardinalities, num_classes, d_token=192, n_blocks=6, attention_dropout=0.2, ffn_dropout=0.2):
7
- super(FTTransformer, self).__init__()
8
-
9
- self.num_classes = num_classes # 클래스 개수 저장
10
-
11
- # Embedding layers for categorical features
12
- self.cat_embeddings = nn.ModuleList([
13
- nn.Embedding(num_categories, d_token) for num_categories in cat_cardinalities
14
- ])
15
-
16
- # Linear layer for numerical features
17
- self.num_linear = nn.Linear(num_features, d_token)
18
-
19
- # Transformer blocks
20
- self.transformer_blocks = nn.ModuleList([
21
- nn.TransformerEncoderLayer(
22
- d_model=d_token,
23
- nhead=8,
24
- dim_feedforward=4 * d_token,
25
- dropout=attention_dropout,
26
- activation='gelu'
27
- ) for _ in range(n_blocks)
28
- ])
29
-
30
- self.ffn_dropout = nn.Dropout(ffn_dropout)
31
- if num_classes == 2:
32
- self.output_layer = nn.Linear(d_token, 1) # Binary classification
33
- elif num_classes > 2:
34
- self.output_layer = nn.Linear(d_token, num_classes) # Multi classification
35
-
36
- def forward(self, x_num, x_cat):
37
- # Numerical feature embedding
38
- x_num = self.num_linear(x_num)
39
-
40
- # Categorical feature embedding
41
- x_cat = [embed(x_cat[:, i]) for i, embed in enumerate(self.cat_embeddings)]
42
- x_cat = torch.stack(x_cat, dim=1)
43
-
44
- # Combine numerical and categorical embeddings
45
- x = x_num.unsqueeze(1) + x_cat
46
-
47
- # Pass through transformer blocks
48
- for block in self.transformer_blocks:
49
- x = block(x)
50
-
51
- # Pooling and output
52
- x = x.mean(dim=1)
53
- x = self.ffn_dropout(x)
54
- x = self.output_layer(x)
55
-
56
- return x
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/make_oversample_data/gan_sample_10000_1.py DELETED
@@ -1,181 +0,0 @@
1
- import pandas as pd
2
- import numpy as np
3
-
4
- from imblearn.over_sampling import SMOTENC
5
- import optuna
6
- from ctgan import CTGAN
7
- import torch
8
- import warnings
9
-
10
- # 지역별 데이터 파일 경로
11
- regions = ['incheon', 'seoul','busan', 'daegu', 'daejeon', 'gwangju']
12
- file_paths = [f'../../data/data_for_modeling/{region}_train.csv' for region in regions]
13
- output_paths = [f'../../data/data_oversampled/ctgan10000/ctgan10000_1_{region}.csv' for region in regions]
14
-
15
- # GPU 사용 설정
16
- device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
17
- print(f"Using device: {device}")
18
-
19
- # 경고 무시
20
- warnings.filterwarnings("ignore", category=UserWarning, module="optuna.distributions")
21
-
22
- # 지역별 처리
23
- for file_path, output_path in zip(file_paths, output_paths):
24
- # 데이터 로드
25
- data = pd.read_csv(file_path, index_col=0)
26
- data= data.loc[data['year'].isin([2018,2019]),:]
27
- data['cloudcover'] = data['cloudcover'].astype('int')
28
- data['lm_cloudcover'] = data['lm_cloudcover'].astype('int')
29
- X = data.drop(columns=['multi_class', 'binary_class'])
30
- y = data['multi_class']
31
-
32
- # 불필요한 열 제거
33
- X.drop(columns=['ground_temp - temp_C', 'hour_sin', 'hour_cos', 'month_sin', 'month_cos'], inplace=True)
34
-
35
- # SMOTENC에서 사용할 범주형 변수 열 번호 설정
36
- categorical_features_indices = [i for i, dtype in enumerate(X.dtypes) if dtype != 'float64']
37
-
38
- # sampling_strategy 설정
39
- count_class_0 = (y == 0).sum()
40
- count_class_1 = (y == 1).sum()
41
- count_class_2 = (y == 2).sum()
42
- sampling_strategy = {
43
- 0: 500 if count_class_0 <= 500 else 1000,
44
- 1: int(np.ceil(count_class_1 / 100) * 100), # 백의 자리로 올림
45
- 2: count_class_2
46
- }
47
-
48
- # SMOTENC 적용
49
- smotenc = SMOTENC(categorical_features=categorical_features_indices, sampling_strategy=sampling_strategy, random_state=42)
50
- X_resampled, y_resampled = smotenc.fit_resample(X, y)
51
-
52
- # Resampled 데이터 생성
53
- lerp_data = X_resampled.copy()
54
- lerp_data['multi_class'] = y_resampled
55
-
56
- # CTGAN에서 사용할 범주형 변수 열 이름 설정
57
- categorical_features = [
58
- col for col, dtype in zip(lerp_data.columns, lerp_data.dtypes) if dtype != 'float64'
59
- ]
60
-
61
- # Optuna 목적 함수 정의
62
- def objective(trial):
63
- # 하이퍼파라미터 탐색 범위 설정
64
- embedding_dim = trial.suggest_int("embedding_dim", 64, 128)
65
- generator_dim = trial.suggest_categorical("generator_dim", [(64, 64), (128, 128)])
66
- discriminator_dim = trial.suggest_categorical("discriminator_dim", [(64, 64), (128, 128)])
67
- pac = trial.suggest_categorical("pac", [4, 8])
68
- batch_size = trial.suggest_categorical("batch_size", [64, 128, 256])
69
- discriminator_steps = trial.suggest_int("discriminator_steps", 1, 3)
70
-
71
- # CTGAN 모델 생성
72
- ctgan = CTGAN(
73
- embedding_dim=embedding_dim,
74
- generator_dim=generator_dim,
75
- discriminator_dim=discriminator_dim,
76
- batch_size=batch_size,
77
- discriminator_steps=discriminator_steps,
78
- pac=pac
79
- )
80
-
81
- # 범주 0 데이터 필터링
82
- data_0 = lerp_data[lerp_data['multi_class'] == 0]
83
-
84
- # 모델 학습
85
- ctgan.fit(data_0, discrete_columns=categorical_features)
86
-
87
- # 샘플 생성
88
- generated_data = ctgan.sample(len(data_0) * 2)
89
-
90
- # 평가: 샘플의 연속형 변수 분포 비교
91
- real_visi = data_0['visi']
92
- generated_visi = generated_data['visi']
93
-
94
- # 분포 간 차이(MSE) 계산
95
- mse = ((real_visi.mean() - generated_visi.mean())**2 + (real_visi.std() - generated_visi.std())**2)
96
- return -mse
97
-
98
- # Optuna로 최적화 수행
99
- study = optuna.create_study(direction="maximize")
100
- study.optimize(objective, n_trials=50)
101
-
102
- # 최적 하이퍼파라미터 출력
103
- best_params = study.best_params
104
-
105
- # 최적 하이퍼파라미터로 CTGAN 학습 및 샘플 생성
106
- ctgan = CTGAN(
107
- embedding_dim=best_params["embedding_dim"],
108
- generator_dim=best_params["generator_dim"],
109
- discriminator_dim=best_params["discriminator_dim"],
110
- batch_size=best_params["batch_size"],
111
- discriminator_steps=best_params["discriminator_steps"],
112
- pac=best_params["pac"]
113
- )
114
-
115
- # 범주 0 데이터로 최종 학습
116
- ctgan.fit(lerp_data[lerp_data['multi_class'] == 0], discrete_columns=categorical_features)
117
- generated_0 = ctgan.sample(10000)
118
-
119
- # 범주 1 데이터 최적화 및 생성
120
- def objective_class1(trial):
121
- embedding_dim = trial.suggest_int("embedding_dim", 128, 512)
122
- generator_dim = trial.suggest_categorical("generator_dim", [(128, 128), (256, 256)])
123
- discriminator_dim = trial.suggest_categorical("discriminator_dim", [(128, 128), (256, 256)])
124
- pac = trial.suggest_categorical("pac", [4, 8])
125
- batch_size = trial.suggest_categorical("batch_size", [256, 512, 1024])
126
- discriminator_steps = trial.suggest_int("discriminator_steps", 1, 5)
127
-
128
- ctgan = CTGAN(
129
- embedding_dim=embedding_dim,
130
- generator_dim=generator_dim,
131
- discriminator_dim=discriminator_dim,
132
- batch_size=batch_size,
133
- discriminator_steps=discriminator_steps,
134
- pac=pac
135
- )
136
-
137
- data_1 = lerp_data[lerp_data['multi_class'] == 1]
138
- ctgan.fit(data_1, discrete_columns=categorical_features)
139
- generated_data = ctgan.sample(len(data_1) * 2)
140
-
141
- real_visi = data_1['visi']
142
- generated_visi = generated_data['visi']
143
- mse = ((real_visi.mean() - generated_visi.mean())**2 + (real_visi.std() - generated_visi.std())**2)
144
- return -mse
145
-
146
- study_class1 = optuna.create_study(direction="maximize")
147
- study_class1.optimize(objective_class1, n_trials=30)
148
-
149
- best_params_class1 = study_class1.best_params
150
- ctgan = CTGAN(
151
- embedding_dim=best_params_class1["embedding_dim"],
152
- generator_dim=best_params_class1["generator_dim"],
153
- discriminator_dim=best_params_class1["discriminator_dim"],
154
- batch_size=best_params_class1["batch_size"],
155
- discriminator_steps=best_params_class1["discriminator_steps"],
156
- pac=best_params_class1["pac"]
157
- )
158
-
159
- ctgan.fit(lerp_data[lerp_data['multi_class'] == 1], discrete_columns=categorical_features)
160
- generated_1 = ctgan.sample(10000 - int(np.ceil(count_class_1 / 100) * 100))
161
-
162
- # 데이터 병합 및 저장
163
- well_generated0 = generated_0[(generated_0['visi'] >= 0) & (generated_0['visi'] < 100)]
164
- well_generated1 = generated_1[(generated_1['visi'] >= 100) & (generated_1['visi'] < 500)]
165
- smote_gan_data = pd.concat([lerp_data, well_generated0, well_generated1], axis=0)
166
- # 제거변수 복구
167
- smote_gan_data['binary_class'] = smote_gan_data['multi_class'].apply(lambda x: 0 if x == 2 else 1)
168
- smote_gan_data['hour_sin'] = np.sin(2 * np.pi * smote_gan_data['hour'] / 24)
169
- smote_gan_data['hour_cos'] = np.cos(2 * np.pi * smote_gan_data['hour'] / 24)
170
- smote_gan_data['month_sin'] = np.sin(2 * np.pi * smote_gan_data['month'] / 12)
171
- smote_gan_data['month_cos'] = np.cos(2 * np.pi * smote_gan_data['month'] / 12)
172
- smote_gan_data['ground_temp - temp_C'] = smote_gan_data['groundtemp'] - smote_gan_data['temp_C']
173
-
174
- filtered_data = smote_gan_data[smote_gan_data['multi_class'] != 2]
175
- original_class2 = data[data['multi_class'] == 2]
176
- final_data = pd.concat([filtered_data, original_class2], axis=0)
177
- final_data.reset_index(drop=True, inplace=True)
178
-
179
- # 결과 저장
180
- final_data.to_csv(output_path, index = False)
181
- print(len(final_data[final_data['multi_class']==0]),'|',len(final_data[final_data['multi_class']==1]),'|',len(final_data[final_data['multi_class']==2]))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/make_oversample_data/gan_sample_10000_2.py DELETED
@@ -1,182 +0,0 @@
1
- import pandas as pd
2
- import numpy as np
3
- import os
4
- os.environ["CUDA_VISIBLE_DEVICES"] = "1"
5
- from imblearn.over_sampling import SMOTENC
6
- import optuna
7
- from ctgan import CTGAN
8
- import torch
9
- import warnings
10
-
11
- # 지역별 데이터 파일 경로
12
- regions = ['incheon', 'seoul','busan', 'daegu', 'daejeon', 'gwangju']
13
- file_paths = [f'../../data/data_for_modeling/{region}_train.csv' for region in regions]
14
- output_paths = [f'../../data/data_oversampled/ctgan10000/ctgan10000_2_{region}.csv' for region in regions]
15
-
16
- # GPU 사용 설정
17
- device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
18
- print(f"Using device: {device}")
19
-
20
- # 경고 무시
21
- warnings.filterwarnings("ignore", category=UserWarning, module="optuna.distributions")
22
-
23
- # 지역별 처리
24
- for file_path, output_path in zip(file_paths, output_paths):
25
- # 데이터 로드
26
- data = pd.read_csv(file_path, index_col=0)
27
- data= data.loc[data['year'].isin([2018,2020]),:]
28
- data['cloudcover'] = data['cloudcover'].astype('int')
29
- data['lm_cloudcover'] = data['lm_cloudcover'].astype('int')
30
- X = data.drop(columns=['multi_class', 'binary_class'])
31
- y = data['multi_class']
32
-
33
- # 불필요한 열 제거
34
- X.drop(columns=['ground_temp - temp_C', 'hour_sin', 'hour_cos', 'month_sin', 'month_cos'], inplace=True)
35
-
36
- # SMOTENC에서 사용할 범주형 변수 열 번호 설정
37
- categorical_features_indices = [i for i, dtype in enumerate(X.dtypes) if dtype != 'float64']
38
-
39
- # sampling_strategy 설정
40
- count_class_0 = (y == 0).sum()
41
- count_class_1 = (y == 1).sum()
42
- count_class_2 = (y == 2).sum()
43
- sampling_strategy = {
44
- 0: 500 if count_class_0 <= 500 else 1000,
45
- 1: int(np.ceil(count_class_1 / 100) * 100), # 백의 자리로 올림
46
- 2: count_class_2
47
- }
48
-
49
- # SMOTENC 적용
50
- smotenc = SMOTENC(categorical_features=categorical_features_indices, sampling_strategy=sampling_strategy, random_state=42)
51
- X_resampled, y_resampled = smotenc.fit_resample(X, y)
52
-
53
- # Resampled 데이터 생성
54
- lerp_data = X_resampled.copy()
55
- lerp_data['multi_class'] = y_resampled
56
-
57
- # CTGAN에서 사용할 범주형 변수 열 이름 설정
58
- categorical_features = [
59
- col for col, dtype in zip(lerp_data.columns, lerp_data.dtypes) if dtype != 'float64'
60
- ]
61
-
62
- # Optuna 목적 함수 정의
63
- def objective(trial):
64
- # 하이퍼파라미터 탐색 범위 설정
65
- embedding_dim = trial.suggest_int("embedding_dim", 64, 128)
66
- generator_dim = trial.suggest_categorical("generator_dim", [(64, 64), (128, 128)])
67
- discriminator_dim = trial.suggest_categorical("discriminator_dim", [(64, 64), (128, 128)])
68
- pac = trial.suggest_categorical("pac", [4, 8])
69
- batch_size = trial.suggest_categorical("batch_size", [64, 128, 256])
70
- discriminator_steps = trial.suggest_int("discriminator_steps", 1, 3)
71
-
72
- # CTGAN 모델 생성
73
- ctgan = CTGAN(
74
- embedding_dim=embedding_dim,
75
- generator_dim=generator_dim,
76
- discriminator_dim=discriminator_dim,
77
- batch_size=batch_size,
78
- discriminator_steps=discriminator_steps,
79
- pac=pac
80
- )
81
-
82
- # 범주 0 데이터 필터링
83
- data_0 = lerp_data[lerp_data['multi_class'] == 0]
84
-
85
- # 모델 학습
86
- ctgan.fit(data_0, discrete_columns=categorical_features)
87
-
88
- # 샘플 생성
89
- generated_data = ctgan.sample(len(data_0) * 2)
90
-
91
- # 평가: 샘플의 연속형 변수 분포 비교
92
- real_visi = data_0['visi']
93
- generated_visi = generated_data['visi']
94
-
95
- # 분포 간 차이(MSE) 계산
96
- mse = ((real_visi.mean() - generated_visi.mean())**2 + (real_visi.std() - generated_visi.std())**2)
97
- return -mse
98
-
99
- # Optuna로 최적화 수행
100
- study = optuna.create_study(direction="maximize")
101
- study.optimize(objective, n_trials=50)
102
-
103
- # 최적 하이퍼파라미터 출력
104
- best_params = study.best_params
105
-
106
- # 최적 하이퍼파라미터로 CTGAN 학습 및 샘플 생성
107
- ctgan = CTGAN(
108
- embedding_dim=best_params["embedding_dim"],
109
- generator_dim=best_params["generator_dim"],
110
- discriminator_dim=best_params["discriminator_dim"],
111
- batch_size=best_params["batch_size"],
112
- discriminator_steps=best_params["discriminator_steps"],
113
- pac=best_params["pac"]
114
- )
115
-
116
- # 범주 0 데이터로 최종 학습
117
- ctgan.fit(lerp_data[lerp_data['multi_class'] == 0], discrete_columns=categorical_features)
118
- generated_0 = ctgan.sample(10000)
119
-
120
- # 범주 1 데이터 최적화 및 생성
121
- def objective_class1(trial):
122
- embedding_dim = trial.suggest_int("embedding_dim", 128, 512)
123
- generator_dim = trial.suggest_categorical("generator_dim", [(128, 128), (256, 256)])
124
- discriminator_dim = trial.suggest_categorical("discriminator_dim", [(128, 128), (256, 256)])
125
- pac = trial.suggest_categorical("pac", [4, 8])
126
- batch_size = trial.suggest_categorical("batch_size", [256, 512, 1024])
127
- discriminator_steps = trial.suggest_int("discriminator_steps", 1, 5)
128
-
129
- ctgan = CTGAN(
130
- embedding_dim=embedding_dim,
131
- generator_dim=generator_dim,
132
- discriminator_dim=discriminator_dim,
133
- batch_size=batch_size,
134
- discriminator_steps=discriminator_steps,
135
- pac=pac
136
- )
137
-
138
- data_1 = lerp_data[lerp_data['multi_class'] == 1]
139
- ctgan.fit(data_1, discrete_columns=categorical_features)
140
- generated_data = ctgan.sample(len(data_1) * 2)
141
-
142
- real_visi = data_1['visi']
143
- generated_visi = generated_data['visi']
144
- mse = ((real_visi.mean() - generated_visi.mean())**2 + (real_visi.std() - generated_visi.std())**2)
145
- return -mse
146
-
147
- study_class1 = optuna.create_study(direction="maximize")
148
- study_class1.optimize(objective_class1, n_trials=30)
149
-
150
- best_params_class1 = study_class1.best_params
151
- ctgan = CTGAN(
152
- embedding_dim=best_params_class1["embedding_dim"],
153
- generator_dim=best_params_class1["generator_dim"],
154
- discriminator_dim=best_params_class1["discriminator_dim"],
155
- batch_size=best_params_class1["batch_size"],
156
- discriminator_steps=best_params_class1["discriminator_steps"],
157
- pac=best_params_class1["pac"]
158
- )
159
-
160
- ctgan.fit(lerp_data[lerp_data['multi_class'] == 1], discrete_columns=categorical_features)
161
- generated_1 = ctgan.sample(10000 - int(np.ceil(count_class_1 / 100) * 100))
162
-
163
- # 데이터 병합 및 저장
164
- well_generated0 = generated_0[(generated_0['visi'] >= 0) & (generated_0['visi'] < 100)]
165
- well_generated1 = generated_1[(generated_1['visi'] >= 100) & (generated_1['visi'] < 500)]
166
- smote_gan_data = pd.concat([lerp_data, well_generated0, well_generated1], axis=0)
167
- # 제거변수 복구
168
- smote_gan_data['binary_class'] = smote_gan_data['multi_class'].apply(lambda x: 0 if x == 2 else 1)
169
- smote_gan_data['hour_sin'] = np.sin(2 * np.pi * smote_gan_data['hour'] / 24)
170
- smote_gan_data['hour_cos'] = np.cos(2 * np.pi * smote_gan_data['hour'] / 24)
171
- smote_gan_data['month_sin'] = np.sin(2 * np.pi * smote_gan_data['month'] / 12)
172
- smote_gan_data['month_cos'] = np.cos(2 * np.pi * smote_gan_data['month'] / 12)
173
- smote_gan_data['ground_temp - temp_C'] = smote_gan_data['groundtemp'] - smote_gan_data['temp_C']
174
-
175
- filtered_data = smote_gan_data[smote_gan_data['multi_class'] != 2]
176
- original_class2 = data[data['multi_class'] == 2]
177
- final_data = pd.concat([filtered_data, original_class2], axis=0)
178
- final_data.reset_index(drop=True, inplace=True)
179
-
180
- # 결과 저장
181
- final_data.to_csv(output_path, index = False)
182
- print(len(final_data[final_data['multi_class']==0]),'|',len(final_data[final_data['multi_class']==1]),'|',len(final_data[final_data['multi_class']==2]))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/make_oversample_data/gan_sample_10000_3.py DELETED
@@ -1,182 +0,0 @@
1
- import pandas as pd
2
- import numpy as np
3
- import os
4
- os.environ["CUDA_VISIBLE_DEVICES"] = "0"
5
- from imblearn.over_sampling import SMOTENC
6
- import optuna
7
- from ctgan import CTGAN
8
- import torch
9
- import warnings
10
-
11
- # 지역별 데이터 파일 경로
12
- regions = ['incheon', 'seoul','busan', 'daegu', 'daejeon', 'gwangju']
13
- file_paths = [f'../../data/data_for_modeling/{region}_train.csv' for region in regions]
14
- output_paths = [f'../../data/data_oversampled/ctgan10000/ctgan10000_3_{region}.csv' for region in regions]
15
-
16
- # GPU 사용 설정
17
- device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
18
- print(f"Using device: {device}")
19
-
20
- # 경고 무시
21
- warnings.filterwarnings("ignore", category=UserWarning, module="optuna.distributions")
22
-
23
- # 지역별 처리
24
- for file_path, output_path in zip(file_paths, output_paths):
25
- # 데이터 로드
26
- data = pd.read_csv(file_path, index_col=0)
27
- data= data.loc[data['year'].isin([2019,2020]),:]
28
- data['cloudcover'] = data['cloudcover'].astype('int')
29
- data['lm_cloudcover'] = data['lm_cloudcover'].astype('int')
30
- X = data.drop(columns=['multi_class', 'binary_class'])
31
- y = data['multi_class']
32
-
33
- # 불필요한 열 제거
34
- X.drop(columns=['ground_temp - temp_C', 'hour_sin', 'hour_cos', 'month_sin', 'month_cos'], inplace=True)
35
-
36
- # SMOTENC에서 사용할 범주형 변수 열 번호 설정
37
- categorical_features_indices = [i for i, dtype in enumerate(X.dtypes) if dtype != 'float64']
38
-
39
- # sampling_strategy 설정
40
- count_class_0 = (y == 0).sum()
41
- count_class_1 = (y == 1).sum()
42
- count_class_2 = (y == 2).sum()
43
- sampling_strategy = {
44
- 0: 500 if count_class_0 <= 500 else 1000,
45
- 1: int(np.ceil(count_class_1 / 100) * 100), # 백의 자리로 올림
46
- 2: count_class_2
47
- }
48
-
49
- # SMOTENC 적용
50
- smotenc = SMOTENC(categorical_features=categorical_features_indices, sampling_strategy=sampling_strategy, random_state=42)
51
- X_resampled, y_resampled = smotenc.fit_resample(X, y)
52
-
53
- # Resampled 데이터 생성
54
- lerp_data = X_resampled.copy()
55
- lerp_data['multi_class'] = y_resampled
56
-
57
- # CTGAN에서 사용할 범주형 변수 열 이름 설정
58
- categorical_features = [
59
- col for col, dtype in zip(lerp_data.columns, lerp_data.dtypes) if dtype != 'float64'
60
- ]
61
-
62
- # Optuna 목적 함수 정의
63
- def objective(trial):
64
- # 하이퍼파라미터 탐색 범위 설정
65
- embedding_dim = trial.suggest_int("embedding_dim", 64, 128)
66
- generator_dim = trial.suggest_categorical("generator_dim", [(64, 64), (128, 128)])
67
- discriminator_dim = trial.suggest_categorical("discriminator_dim", [(64, 64), (128, 128)])
68
- pac = trial.suggest_categorical("pac", [4, 8])
69
- batch_size = trial.suggest_categorical("batch_size", [64, 128, 256])
70
- discriminator_steps = trial.suggest_int("discriminator_steps", 1, 3)
71
-
72
- # CTGAN 모델 생성
73
- ctgan = CTGAN(
74
- embedding_dim=embedding_dim,
75
- generator_dim=generator_dim,
76
- discriminator_dim=discriminator_dim,
77
- batch_size=batch_size,
78
- discriminator_steps=discriminator_steps,
79
- pac=pac
80
- )
81
-
82
- # 범주 0 데이터 필터링
83
- data_0 = lerp_data[lerp_data['multi_class'] == 0]
84
-
85
- # 모델 학습
86
- ctgan.fit(data_0, discrete_columns=categorical_features)
87
-
88
- # 샘플 생성
89
- generated_data = ctgan.sample(len(data_0) * 2)
90
-
91
- # 평가: 샘플의 연속형 변수 분포 비교
92
- real_visi = data_0['visi']
93
- generated_visi = generated_data['visi']
94
-
95
- # 분포 간 차이(MSE) 계산
96
- mse = ((real_visi.mean() - generated_visi.mean())**2 + (real_visi.std() - generated_visi.std())**2)
97
- return -mse
98
-
99
- # Optuna로 최적화 수행
100
- study = optuna.create_study(direction="maximize")
101
- study.optimize(objective, n_trials=50)
102
-
103
- # 최적 하이퍼파라미터 출력
104
- best_params = study.best_params
105
-
106
- # 최적 하이퍼파라미터로 CTGAN 학습 및 샘플 생성
107
- ctgan = CTGAN(
108
- embedding_dim=best_params["embedding_dim"],
109
- generator_dim=best_params["generator_dim"],
110
- discriminator_dim=best_params["discriminator_dim"],
111
- batch_size=best_params["batch_size"],
112
- discriminator_steps=best_params["discriminator_steps"],
113
- pac=best_params["pac"]
114
- )
115
-
116
- # 범주 0 데이터로 최종 학습
117
- ctgan.fit(lerp_data[lerp_data['multi_class'] == 0], discrete_columns=categorical_features)
118
- generated_0 = ctgan.sample(10000)
119
-
120
- # 범주 1 데이터 최적화 및 생성
121
- def objective_class1(trial):
122
- embedding_dim = trial.suggest_int("embedding_dim", 128, 512)
123
- generator_dim = trial.suggest_categorical("generator_dim", [(128, 128), (256, 256)])
124
- discriminator_dim = trial.suggest_categorical("discriminator_dim", [(128, 128), (256, 256)])
125
- pac = trial.suggest_categorical("pac", [4, 8])
126
- batch_size = trial.suggest_categorical("batch_size", [256, 512, 1024])
127
- discriminator_steps = trial.suggest_int("discriminator_steps", 1, 5)
128
-
129
- ctgan = CTGAN(
130
- embedding_dim=embedding_dim,
131
- generator_dim=generator_dim,
132
- discriminator_dim=discriminator_dim,
133
- batch_size=batch_size,
134
- discriminator_steps=discriminator_steps,
135
- pac=pac
136
- )
137
-
138
- data_1 = lerp_data[lerp_data['multi_class'] == 1]
139
- ctgan.fit(data_1, discrete_columns=categorical_features)
140
- generated_data = ctgan.sample(len(data_1) * 2)
141
-
142
- real_visi = data_1['visi']
143
- generated_visi = generated_data['visi']
144
- mse = ((real_visi.mean() - generated_visi.mean())**2 + (real_visi.std() - generated_visi.std())**2)
145
- return -mse
146
-
147
- study_class1 = optuna.create_study(direction="maximize")
148
- study_class1.optimize(objective_class1, n_trials=30)
149
-
150
- best_params_class1 = study_class1.best_params
151
- ctgan = CTGAN(
152
- embedding_dim=best_params_class1["embedding_dim"],
153
- generator_dim=best_params_class1["generator_dim"],
154
- discriminator_dim=best_params_class1["discriminator_dim"],
155
- batch_size=best_params_class1["batch_size"],
156
- discriminator_steps=best_params_class1["discriminator_steps"],
157
- pac=best_params_class1["pac"]
158
- )
159
-
160
- ctgan.fit(lerp_data[lerp_data['multi_class'] == 1], discrete_columns=categorical_features)
161
- generated_1 = ctgan.sample(10000 - int(np.ceil(count_class_1 / 100) * 100))
162
-
163
- # 데이터 병합 및 저장
164
- well_generated0 = generated_0[(generated_0['visi'] >= 0) & (generated_0['visi'] < 100)]
165
- well_generated1 = generated_1[(generated_1['visi'] >= 100) & (generated_1['visi'] < 500)]
166
- smote_gan_data = pd.concat([lerp_data, well_generated0, well_generated1], axis=0)
167
- # 제거변수 복구
168
- smote_gan_data['binary_class'] = smote_gan_data['multi_class'].apply(lambda x: 0 if x == 2 else 1)
169
- smote_gan_data['hour_sin'] = np.sin(2 * np.pi * smote_gan_data['hour'] / 24)
170
- smote_gan_data['hour_cos'] = np.cos(2 * np.pi * smote_gan_data['hour'] / 24)
171
- smote_gan_data['month_sin'] = np.sin(2 * np.pi * smote_gan_data['month'] / 12)
172
- smote_gan_data['month_cos'] = np.cos(2 * np.pi * smote_gan_data['month'] / 12)
173
- smote_gan_data['ground_temp - temp_C'] = smote_gan_data['groundtemp'] - smote_gan_data['temp_C']
174
-
175
- filtered_data = smote_gan_data[smote_gan_data['multi_class'] != 2]
176
- original_class2 = data[data['multi_class'] == 2]
177
- final_data = pd.concat([filtered_data, original_class2], axis=0)
178
- final_data.reset_index(drop=True, inplace=True)
179
-
180
- # 결과 저장
181
- final_data.to_csv(output_path, index = False)
182
- print(len(final_data[final_data['multi_class']==0]),'|',len(final_data[final_data['multi_class']==1]),'|',len(final_data[final_data['multi_class']==2]))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/make_oversample_data/gan_sample_20000_1.py DELETED
@@ -1,183 +0,0 @@
1
- import pandas as pd
2
- import numpy as np
3
- # import os
4
- # os.environ["CUDA_VISIBLE_DEVICES"] = "1"
5
- from imblearn.over_sampling import SMOTENC
6
- import optuna
7
- from ctgan import CTGAN
8
- import torch
9
- import warnings
10
-
11
- # 지역별 데이터 파일 경로
12
- # regions = ['busan', 'daegu', 'daejeon', 'gwangju', 'incheon', 'seoul']
13
- regions = ['incheon', 'seoul','busan', 'daegu', 'daejeon', 'gwangju']
14
- file_paths = [f'../../data/data_for_modeling/{region}_train.csv' for region in regions]
15
- output_paths = [f'../../data/data_oversampled/ctgan20000/ctgan20000_1_{region}.csv' for region in regions]
16
-
17
- # GPU 사용 설정
18
- device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
19
- print(f"Using device: {device}")
20
-
21
- # 경고 무시
22
- warnings.filterwarnings("ignore", category=UserWarning, module="optuna.distributions")
23
-
24
- # 지역별 처리
25
- for file_path, output_path in zip(file_paths, output_paths):
26
- # 데이터 로드
27
- data = pd.read_csv(file_path, index_col=0)
28
- data= data.loc[data['year'].isin([2018,2019]),:]
29
- data['cloudcover'] = data['cloudcover'].astype('int')
30
- data['lm_cloudcover'] = data['lm_cloudcover'].astype('int')
31
- X = data.drop(columns=['multi_class', 'binary_class'])
32
- y = data['multi_class']
33
-
34
- # 불필요한 열 제거
35
- X.drop(columns=['ground_temp - temp_C', 'hour_sin', 'hour_cos', 'month_sin', 'month_cos'], inplace=True)
36
-
37
- # SMOTENC에서 사용할 범주형 변수 열 번호 설정
38
- categorical_features_indices = [i for i, dtype in enumerate(X.dtypes) if dtype != 'float64']
39
-
40
- # sampling_strategy 설정
41
- count_class_0 = (y == 0).sum()
42
- count_class_1 = (y == 1).sum()
43
- count_class_2 = (y == 2).sum()
44
- sampling_strategy = {
45
- 0: 500 if count_class_0 <= 500 else 1000,
46
- 1: int(np.ceil(count_class_1 / 100) * 100), # 백의 자리로 올림
47
- 2: count_class_2
48
- }
49
-
50
- # SMOTENC 적용
51
- smotenc = SMOTENC(categorical_features=categorical_features_indices, sampling_strategy=sampling_strategy, random_state=42)
52
- X_resampled, y_resampled = smotenc.fit_resample(X, y)
53
-
54
- # Resampled 데이터 생성
55
- lerp_data = X_resampled.copy()
56
- lerp_data['multi_class'] = y_resampled
57
-
58
- # CTGAN에서 사용할 범주형 변수 열 이름 설정
59
- categorical_features = [
60
- col for col, dtype in zip(lerp_data.columns, lerp_data.dtypes) if dtype != 'float64'
61
- ]
62
-
63
- # Optuna 목적 함수 정의
64
- def objective(trial):
65
- # 하이퍼파라미터 탐색 범위 설정
66
- embedding_dim = trial.suggest_int("embedding_dim", 64, 128)
67
- generator_dim = trial.suggest_categorical("generator_dim", [(64, 64), (128, 128)])
68
- discriminator_dim = trial.suggest_categorical("discriminator_dim", [(64, 64), (128, 128)])
69
- pac = trial.suggest_categorical("pac", [4, 8])
70
- batch_size = trial.suggest_categorical("batch_size", [64, 128, 256])
71
- discriminator_steps = trial.suggest_int("discriminator_steps", 1, 3)
72
-
73
- # CTGAN 모델 생성
74
- ctgan = CTGAN(
75
- embedding_dim=embedding_dim,
76
- generator_dim=generator_dim,
77
- discriminator_dim=discriminator_dim,
78
- batch_size=batch_size,
79
- discriminator_steps=discriminator_steps,
80
- pac=pac
81
- )
82
-
83
- # 범주 0 데이터 필터링
84
- data_0 = lerp_data[lerp_data['multi_class'] == 0]
85
-
86
- # 모델 학습
87
- ctgan.fit(data_0, discrete_columns=categorical_features)
88
-
89
- # 샘플 생성
90
- generated_data = ctgan.sample(len(data_0) * 2)
91
-
92
- # 평가: 샘플의 연속형 변수 분포 비교
93
- real_visi = data_0['visi']
94
- generated_visi = generated_data['visi']
95
-
96
- # 분포 간 차이(MSE) 계산
97
- mse = ((real_visi.mean() - generated_visi.mean())**2 + (real_visi.std() - generated_visi.std())**2)
98
- return -mse
99
-
100
- # Optuna로 최적화 수행
101
- study = optuna.create_study(direction="maximize")
102
- study.optimize(objective, n_trials=50)
103
-
104
- # 최적 하이퍼파라미터 출력
105
- best_params = study.best_params
106
-
107
- # 최적 하이퍼파라미터로 CTGAN 학습 및 샘플 생성
108
- ctgan = CTGAN(
109
- embedding_dim=best_params["embedding_dim"],
110
- generator_dim=best_params["generator_dim"],
111
- discriminator_dim=best_params["discriminator_dim"],
112
- batch_size=best_params["batch_size"],
113
- discriminator_steps=best_params["discriminator_steps"],
114
- pac=best_params["pac"]
115
- )
116
-
117
- # 범주 0 데이터로 최종 학습
118
- ctgan.fit(lerp_data[lerp_data['multi_class'] == 0], discrete_columns=categorical_features)
119
- generated_0 = ctgan.sample(20000)
120
-
121
- # 범주 1 데이터 최적화 및 생성
122
- def objective_class1(trial):
123
- embedding_dim = trial.suggest_int("embedding_dim", 128, 512)
124
- generator_dim = trial.suggest_categorical("generator_dim", [(128, 128), (256, 256)])
125
- discriminator_dim = trial.suggest_categorical("discriminator_dim", [(128, 128), (256, 256)])
126
- pac = trial.suggest_categorical("pac", [4, 8])
127
- batch_size = trial.suggest_categorical("batch_size", [256, 512, 1024])
128
- discriminator_steps = trial.suggest_int("discriminator_steps", 1, 5)
129
-
130
- ctgan = CTGAN(
131
- embedding_dim=embedding_dim,
132
- generator_dim=generator_dim,
133
- discriminator_dim=discriminator_dim,
134
- batch_size=batch_size,
135
- discriminator_steps=discriminator_steps,
136
- pac=pac
137
- )
138
-
139
- data_1 = lerp_data[lerp_data['multi_class'] == 1]
140
- ctgan.fit(data_1, discrete_columns=categorical_features)
141
- generated_data = ctgan.sample(len(data_1) * 2)
142
-
143
- real_visi = data_1['visi']
144
- generated_visi = generated_data['visi']
145
- mse = ((real_visi.mean() - generated_visi.mean())**2 + (real_visi.std() - generated_visi.std())**2)
146
- return -mse
147
-
148
- study_class1 = optuna.create_study(direction="maximize")
149
- study_class1.optimize(objective_class1, n_trials=30)
150
-
151
- best_params_class1 = study_class1.best_params
152
- ctgan = CTGAN(
153
- embedding_dim=best_params_class1["embedding_dim"],
154
- generator_dim=best_params_class1["generator_dim"],
155
- discriminator_dim=best_params_class1["discriminator_dim"],
156
- batch_size=best_params_class1["batch_size"],
157
- discriminator_steps=best_params_class1["discriminator_steps"],
158
- pac=best_params_class1["pac"]
159
- )
160
-
161
- ctgan.fit(lerp_data[lerp_data['multi_class'] == 1], discrete_columns=categorical_features)
162
- generated_1 = ctgan.sample(20000 - int(np.ceil(count_class_1 / 100) * 100))
163
-
164
- # 데이터 병합 및 저장
165
- well_generated0 = generated_0[(generated_0['visi'] >= 0) & (generated_0['visi'] < 100)]
166
- well_generated1 = generated_1[(generated_1['visi'] >= 100) & (generated_1['visi'] < 500)]
167
- smote_gan_data = pd.concat([lerp_data, well_generated0, well_generated1], axis=0)
168
- # 제거변수 복구
169
- smote_gan_data['binary_class'] = smote_gan_data['multi_class'].apply(lambda x: 0 if x == 2 else 1)
170
- smote_gan_data['hour_sin'] = np.sin(2 * np.pi * smote_gan_data['hour'] / 24)
171
- smote_gan_data['hour_cos'] = np.cos(2 * np.pi * smote_gan_data['hour'] / 24)
172
- smote_gan_data['month_sin'] = np.sin(2 * np.pi * smote_gan_data['month'] / 12)
173
- smote_gan_data['month_cos'] = np.cos(2 * np.pi * smote_gan_data['month'] / 12)
174
- smote_gan_data['ground_temp - temp_C'] = smote_gan_data['groundtemp'] - smote_gan_data['temp_C']
175
-
176
- filtered_data = smote_gan_data[smote_gan_data['multi_class'] != 2]
177
- original_class2 = data[data['multi_class'] == 2]
178
- final_data = pd.concat([filtered_data, original_class2], axis=0)
179
- final_data.reset_index(drop=True, inplace=True)
180
-
181
- # 결과 저장
182
- final_data.to_csv(output_path, index = False)
183
- print(len(final_data[final_data['multi_class']==0]),'|',len(final_data[final_data['multi_class']==1]),'|',len(final_data[final_data['multi_class']==2]))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/make_oversample_data/gan_sample_20000_2.py DELETED
@@ -1,183 +0,0 @@
1
- import pandas as pd
2
- import numpy as np
3
- import os
4
- os.environ["CUDA_VISIBLE_DEVICES"] = "1"
5
- from imblearn.over_sampling import SMOTENC
6
- import optuna
7
- from ctgan import CTGAN
8
- import torch
9
- import warnings
10
-
11
- # 지역별 데이터 파일 경로
12
- # regions = ['busan', 'daegu', 'daejeon', 'gwangju', 'incheon', 'seoul']
13
- regions = ['incheon', 'seoul','busan', 'daegu', 'daejeon', 'gwangju']
14
- file_paths = [f'../../data/data_for_modeling/{region}_train.csv' for region in regions]
15
- output_paths = [f'../../data/data_oversampled/ctgan20000/ctgan20000_2_{region}.csv' for region in regions]
16
-
17
- # GPU 사용 설정
18
- device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
19
- print(f"Using device: {device}")
20
-
21
- # 경고 무시
22
- warnings.filterwarnings("ignore", category=UserWarning, module="optuna.distributions")
23
-
24
- # 지역별 처리
25
- for file_path, output_path in zip(file_paths, output_paths):
26
- # 데이터 로드
27
- data = pd.read_csv(file_path, index_col=0)
28
- data= data.loc[data['year'].isin([2018,2020]),:]
29
- data['cloudcover'] = data['cloudcover'].astype('int')
30
- data['lm_cloudcover'] = data['lm_cloudcover'].astype('int')
31
- X = data.drop(columns=['multi_class', 'binary_class'])
32
- y = data['multi_class']
33
-
34
- # 불필요한 열 제거
35
- X.drop(columns=['ground_temp - temp_C', 'hour_sin', 'hour_cos', 'month_sin', 'month_cos'], inplace=True)
36
-
37
- # SMOTENC에서 사용할 범주형 변수 열 번호 설정
38
- categorical_features_indices = [i for i, dtype in enumerate(X.dtypes) if dtype != 'float64']
39
-
40
- # sampling_strategy 설정
41
- count_class_0 = (y == 0).sum()
42
- count_class_1 = (y == 1).sum()
43
- count_class_2 = (y == 2).sum()
44
- sampling_strategy = {
45
- 0: 500 if count_class_0 <= 500 else 1000,
46
- 1: int(np.ceil(count_class_1 / 100) * 100), # 백의 자리로 올림
47
- 2: count_class_2
48
- }
49
-
50
- # SMOTENC 적용
51
- smotenc = SMOTENC(categorical_features=categorical_features_indices, sampling_strategy=sampling_strategy, random_state=42)
52
- X_resampled, y_resampled = smotenc.fit_resample(X, y)
53
-
54
- # Resampled 데이터 생성
55
- lerp_data = X_resampled.copy()
56
- lerp_data['multi_class'] = y_resampled
57
-
58
- # CTGAN에서 사용할 범주형 변수 열 이름 설정
59
- categorical_features = [
60
- col for col, dtype in zip(lerp_data.columns, lerp_data.dtypes) if dtype != 'float64'
61
- ]
62
-
63
- # Optuna 목적 함수 정의
64
- def objective(trial):
65
- # 하이퍼파라미터 탐색 범위 설정
66
- embedding_dim = trial.suggest_int("embedding_dim", 64, 128)
67
- generator_dim = trial.suggest_categorical("generator_dim", [(64, 64), (128, 128)])
68
- discriminator_dim = trial.suggest_categorical("discriminator_dim", [(64, 64), (128, 128)])
69
- pac = trial.suggest_categorical("pac", [4, 8])
70
- batch_size = trial.suggest_categorical("batch_size", [64, 128, 256])
71
- discriminator_steps = trial.suggest_int("discriminator_steps", 1, 3)
72
-
73
- # CTGAN 모델 생성
74
- ctgan = CTGAN(
75
- embedding_dim=embedding_dim,
76
- generator_dim=generator_dim,
77
- discriminator_dim=discriminator_dim,
78
- batch_size=batch_size,
79
- discriminator_steps=discriminator_steps,
80
- pac=pac
81
- )
82
-
83
- # 범주 0 데이터 필터링
84
- data_0 = lerp_data[lerp_data['multi_class'] == 0]
85
-
86
- # 모델 학습
87
- ctgan.fit(data_0, discrete_columns=categorical_features)
88
-
89
- # 샘플 생성
90
- generated_data = ctgan.sample(len(data_0) * 2)
91
-
92
- # 평가: 샘플의 연속형 변수 분포 비교
93
- real_visi = data_0['visi']
94
- generated_visi = generated_data['visi']
95
-
96
- # 분포 간 차이(MSE) 계산
97
- mse = ((real_visi.mean() - generated_visi.mean())**2 + (real_visi.std() - generated_visi.std())**2)
98
- return -mse
99
-
100
- # Optuna로 최적화 수행
101
- study = optuna.create_study(direction="maximize")
102
- study.optimize(objective, n_trials=50)
103
-
104
- # 최적 하이퍼파라미터 출력
105
- best_params = study.best_params
106
-
107
- # 최적 하이퍼파라미터로 CTGAN 학습 및 샘플 생성
108
- ctgan = CTGAN(
109
- embedding_dim=best_params["embedding_dim"],
110
- generator_dim=best_params["generator_dim"],
111
- discriminator_dim=best_params["discriminator_dim"],
112
- batch_size=best_params["batch_size"],
113
- discriminator_steps=best_params["discriminator_steps"],
114
- pac=best_params["pac"]
115
- )
116
-
117
- # 범주 0 데이터로 최종 학습
118
- ctgan.fit(lerp_data[lerp_data['multi_class'] == 0], discrete_columns=categorical_features)
119
- generated_0 = ctgan.sample(20000)
120
-
121
- # 범주 1 데이터 최적화 및 생성
122
- def objective_class1(trial):
123
- embedding_dim = trial.suggest_int("embedding_dim", 128, 512)
124
- generator_dim = trial.suggest_categorical("generator_dim", [(128, 128), (256, 256)])
125
- discriminator_dim = trial.suggest_categorical("discriminator_dim", [(128, 128), (256, 256)])
126
- pac = trial.suggest_categorical("pac", [4, 8])
127
- batch_size = trial.suggest_categorical("batch_size", [256, 512, 1024])
128
- discriminator_steps = trial.suggest_int("discriminator_steps", 1, 5)
129
-
130
- ctgan = CTGAN(
131
- embedding_dim=embedding_dim,
132
- generator_dim=generator_dim,
133
- discriminator_dim=discriminator_dim,
134
- batch_size=batch_size,
135
- discriminator_steps=discriminator_steps,
136
- pac=pac
137
- )
138
-
139
- data_1 = lerp_data[lerp_data['multi_class'] == 1]
140
- ctgan.fit(data_1, discrete_columns=categorical_features)
141
- generated_data = ctgan.sample(len(data_1) * 2)
142
-
143
- real_visi = data_1['visi']
144
- generated_visi = generated_data['visi']
145
- mse = ((real_visi.mean() - generated_visi.mean())**2 + (real_visi.std() - generated_visi.std())**2)
146
- return -mse
147
-
148
- study_class1 = optuna.create_study(direction="maximize")
149
- study_class1.optimize(objective_class1, n_trials=30)
150
-
151
- best_params_class1 = study_class1.best_params
152
- ctgan = CTGAN(
153
- embedding_dim=best_params_class1["embedding_dim"],
154
- generator_dim=best_params_class1["generator_dim"],
155
- discriminator_dim=best_params_class1["discriminator_dim"],
156
- batch_size=best_params_class1["batch_size"],
157
- discriminator_steps=best_params_class1["discriminator_steps"],
158
- pac=best_params_class1["pac"]
159
- )
160
-
161
- ctgan.fit(lerp_data[lerp_data['multi_class'] == 1], discrete_columns=categorical_features)
162
- generated_1 = ctgan.sample(20000 - int(np.ceil(count_class_1 / 100) * 100))
163
-
164
- # 데이터 병합 및 저장
165
- well_generated0 = generated_0[(generated_0['visi'] >= 0) & (generated_0['visi'] < 100)]
166
- well_generated1 = generated_1[(generated_1['visi'] >= 100) & (generated_1['visi'] < 500)]
167
- smote_gan_data = pd.concat([lerp_data, well_generated0, well_generated1], axis=0)
168
- # 제거변수 복구
169
- smote_gan_data['binary_class'] = smote_gan_data['multi_class'].apply(lambda x: 0 if x == 2 else 1)
170
- smote_gan_data['hour_sin'] = np.sin(2 * np.pi * smote_gan_data['hour'] / 24)
171
- smote_gan_data['hour_cos'] = np.cos(2 * np.pi * smote_gan_data['hour'] / 24)
172
- smote_gan_data['month_sin'] = np.sin(2 * np.pi * smote_gan_data['month'] / 12)
173
- smote_gan_data['month_cos'] = np.cos(2 * np.pi * smote_gan_data['month'] / 12)
174
- smote_gan_data['ground_temp - temp_C'] = smote_gan_data['groundtemp'] - smote_gan_data['temp_C']
175
-
176
- filtered_data = smote_gan_data[smote_gan_data['multi_class'] != 2]
177
- original_class2 = data[data['multi_class'] == 2]
178
- final_data = pd.concat([filtered_data, original_class2], axis=0)
179
- final_data.reset_index(drop=True, inplace=True)
180
-
181
- # 결과 저장
182
- final_data.to_csv(output_path, index = False)
183
- print(len(final_data[final_data['multi_class']==0]),'|',len(final_data[final_data['multi_class']==1]),'|',len(final_data[final_data['multi_class']==2]))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/make_oversample_data/gan_sample_20000_3.py DELETED
@@ -1,183 +0,0 @@
1
- import pandas as pd
2
- import numpy as np
3
- import os
4
- os.environ["CUDA_VISIBLE_DEVICES"] = "1"
5
- from imblearn.over_sampling import SMOTENC
6
- import optuna
7
- from ctgan import CTGAN
8
- import torch
9
- import warnings
10
-
11
- # 지역별 데이터 파일 경로
12
- # regions = ['busan', 'daegu', 'daejeon', 'gwangju', 'incheon', 'seoul']
13
- regions = ['incheon', 'seoul','busan', 'daegu', 'daejeon', 'gwangju']
14
- file_paths = [f'../../data/data_for_modeling/{region}_train.csv' for region in regions]
15
- output_paths = [f'../../data/data_oversampled/ctgan20000/ctgan20000_3_{region}.csv' for region in regions]
16
-
17
- # GPU 사용 설정
18
- device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
19
- print(f"Using device: {device}")
20
-
21
- # 경고 무시
22
- warnings.filterwarnings("ignore", category=UserWarning, module="optuna.distributions")
23
-
24
- # 지역별 처리
25
- for file_path, output_path in zip(file_paths, output_paths):
26
- # 데이터 로드
27
- data = pd.read_csv(file_path, index_col=0)
28
- data= data.loc[data['year'].isin([2019,2020]),:]
29
- data['cloudcover'] = data['cloudcover'].astype('int')
30
- data['lm_cloudcover'] = data['lm_cloudcover'].astype('int')
31
- X = data.drop(columns=['multi_class', 'binary_class'])
32
- y = data['multi_class']
33
-
34
- # 불필요한 열 제거
35
- X.drop(columns=['ground_temp - temp_C', 'hour_sin', 'hour_cos', 'month_sin', 'month_cos'], inplace=True)
36
-
37
- # SMOTENC에서 사용할 범주형 변수 열 번호 설정
38
- categorical_features_indices = [i for i, dtype in enumerate(X.dtypes) if dtype != 'float64']
39
-
40
- # sampling_strategy 설정
41
- count_class_0 = (y == 0).sum()
42
- count_class_1 = (y == 1).sum()
43
- count_class_2 = (y == 2).sum()
44
- sampling_strategy = {
45
- 0: 500 if count_class_0 <= 500 else 1000,
46
- 1: int(np.ceil(count_class_1 / 100) * 100), # 백의 자리로 올림
47
- 2: count_class_2
48
- }
49
-
50
- # SMOTENC 적용
51
- smotenc = SMOTENC(categorical_features=categorical_features_indices, sampling_strategy=sampling_strategy, random_state=42)
52
- X_resampled, y_resampled = smotenc.fit_resample(X, y)
53
-
54
- # Resampled 데이터 생성
55
- lerp_data = X_resampled.copy()
56
- lerp_data['multi_class'] = y_resampled
57
-
58
- # CTGAN에서 사용할 범주형 변수 열 이름 설정
59
- categorical_features = [
60
- col for col, dtype in zip(lerp_data.columns, lerp_data.dtypes) if dtype != 'float64'
61
- ]
62
-
63
- # Optuna 목적 함수 정의
64
- def objective(trial):
65
- # 하이퍼파라미터 탐색 범위 설정
66
- embedding_dim = trial.suggest_int("embedding_dim", 64, 128)
67
- generator_dim = trial.suggest_categorical("generator_dim", [(64, 64), (128, 128)])
68
- discriminator_dim = trial.suggest_categorical("discriminator_dim", [(64, 64), (128, 128)])
69
- pac = trial.suggest_categorical("pac", [4, 8])
70
- batch_size = trial.suggest_categorical("batch_size", [64, 128, 256])
71
- discriminator_steps = trial.suggest_int("discriminator_steps", 1, 3)
72
-
73
- # CTGAN 모델 생성
74
- ctgan = CTGAN(
75
- embedding_dim=embedding_dim,
76
- generator_dim=generator_dim,
77
- discriminator_dim=discriminator_dim,
78
- batch_size=batch_size,
79
- discriminator_steps=discriminator_steps,
80
- pac=pac
81
- )
82
-
83
- # 범주 0 데이터 필터링
84
- data_0 = lerp_data[lerp_data['multi_class'] == 0]
85
-
86
- # 모델 학습
87
- ctgan.fit(data_0, discrete_columns=categorical_features)
88
-
89
- # 샘플 생성
90
- generated_data = ctgan.sample(len(data_0) * 2)
91
-
92
- # 평가: 샘플의 연속형 변수 분포 비교
93
- real_visi = data_0['visi']
94
- generated_visi = generated_data['visi']
95
-
96
- # 분포 간 차이(MSE) 계산
97
- mse = ((real_visi.mean() - generated_visi.mean())**2 + (real_visi.std() - generated_visi.std())**2)
98
- return -mse
99
-
100
- # Optuna로 최적화 수행
101
- study = optuna.create_study(direction="maximize")
102
- study.optimize(objective, n_trials=50)
103
-
104
- # 최적 하이퍼파라미터 출력
105
- best_params = study.best_params
106
-
107
- # 최적 하이퍼파라미터로 CTGAN 학습 및 샘플 생성
108
- ctgan = CTGAN(
109
- embedding_dim=best_params["embedding_dim"],
110
- generator_dim=best_params["generator_dim"],
111
- discriminator_dim=best_params["discriminator_dim"],
112
- batch_size=best_params["batch_size"],
113
- discriminator_steps=best_params["discriminator_steps"],
114
- pac=best_params["pac"]
115
- )
116
-
117
- # 범주 0 데이터로 최종 학습
118
- ctgan.fit(lerp_data[lerp_data['multi_class'] == 0], discrete_columns=categorical_features)
119
- generated_0 = ctgan.sample(20000)
120
-
121
- # 범주 1 데이터 최적화 및 생성
122
- def objective_class1(trial):
123
- embedding_dim = trial.suggest_int("embedding_dim", 128, 512)
124
- generator_dim = trial.suggest_categorical("generator_dim", [(128, 128), (256, 256)])
125
- discriminator_dim = trial.suggest_categorical("discriminator_dim", [(128, 128), (256, 256)])
126
- pac = trial.suggest_categorical("pac", [4, 8])
127
- batch_size = trial.suggest_categorical("batch_size", [256, 512, 1024])
128
- discriminator_steps = trial.suggest_int("discriminator_steps", 1, 5)
129
-
130
- ctgan = CTGAN(
131
- embedding_dim=embedding_dim,
132
- generator_dim=generator_dim,
133
- discriminator_dim=discriminator_dim,
134
- batch_size=batch_size,
135
- discriminator_steps=discriminator_steps,
136
- pac=pac
137
- )
138
-
139
- data_1 = lerp_data[lerp_data['multi_class'] == 1]
140
- ctgan.fit(data_1, discrete_columns=categorical_features)
141
- generated_data = ctgan.sample(len(data_1) * 2)
142
-
143
- real_visi = data_1['visi']
144
- generated_visi = generated_data['visi']
145
- mse = ((real_visi.mean() - generated_visi.mean())**2 + (real_visi.std() - generated_visi.std())**2)
146
- return -mse
147
-
148
- study_class1 = optuna.create_study(direction="maximize")
149
- study_class1.optimize(objective_class1, n_trials=30)
150
-
151
- best_params_class1 = study_class1.best_params
152
- ctgan = CTGAN(
153
- embedding_dim=best_params_class1["embedding_dim"],
154
- generator_dim=best_params_class1["generator_dim"],
155
- discriminator_dim=best_params_class1["discriminator_dim"],
156
- batch_size=best_params_class1["batch_size"],
157
- discriminator_steps=best_params_class1["discriminator_steps"],
158
- pac=best_params_class1["pac"]
159
- )
160
-
161
- ctgan.fit(lerp_data[lerp_data['multi_class'] == 1], discrete_columns=categorical_features)
162
- generated_1 = ctgan.sample(20000 - int(np.ceil(count_class_1 / 100) * 100))
163
-
164
- # 데이터 병합 및 저장
165
- well_generated0 = generated_0[(generated_0['visi'] >= 0) & (generated_0['visi'] < 100)]
166
- well_generated1 = generated_1[(generated_1['visi'] >= 100) & (generated_1['visi'] < 500)]
167
- smote_gan_data = pd.concat([lerp_data, well_generated0, well_generated1], axis=0)
168
- # 제거변수 복구
169
- smote_gan_data['binary_class'] = smote_gan_data['multi_class'].apply(lambda x: 0 if x == 2 else 1)
170
- smote_gan_data['hour_sin'] = np.sin(2 * np.pi * smote_gan_data['hour'] / 24)
171
- smote_gan_data['hour_cos'] = np.cos(2 * np.pi * smote_gan_data['hour'] / 24)
172
- smote_gan_data['month_sin'] = np.sin(2 * np.pi * smote_gan_data['month'] / 12)
173
- smote_gan_data['month_cos'] = np.cos(2 * np.pi * smote_gan_data['month'] / 12)
174
- smote_gan_data['ground_temp - temp_C'] = smote_gan_data['groundtemp'] - smote_gan_data['temp_C']
175
-
176
- filtered_data = smote_gan_data[smote_gan_data['multi_class'] != 2]
177
- original_class2 = data[data['multi_class'] == 2]
178
- final_data = pd.concat([filtered_data, original_class2], axis=0)
179
- final_data.reset_index(drop=True, inplace=True)
180
-
181
- # 결과 저장
182
- final_data.to_csv(output_path, index = False)
183
- print(len(final_data[final_data['multi_class']==0]),'|',len(final_data[final_data['multi_class']==1]),'|',len(final_data[final_data['multi_class']==2]))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/make_oversample_data/gan_sample_7000_1.py DELETED
@@ -1,180 +0,0 @@
1
- import pandas as pd
2
- import numpy as np
3
- from imblearn.over_sampling import SMOTENC
4
- import optuna
5
- from ctgan import CTGAN
6
- import torch
7
- import warnings
8
-
9
- # 지역별 데이터 파일 경로
10
- regions = ['incheon', 'seoul','busan', 'daegu', 'daejeon', 'gwangju']
11
- file_paths = [f'../../data/data_for_modeling/{region}_train.csv' for region in regions]
12
- output_paths = [f'../../data/data_oversampled/ctgan7000/ctgan7000_1_{region}.csv' for region in regions]
13
-
14
- # GPU 사용 설정
15
- device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
16
- print(f"Using device: {device}")
17
-
18
- # 경고 무시
19
- warnings.filterwarnings("ignore", category=UserWarning, module="optuna.distributions")
20
-
21
- # 지역별 처리
22
- for file_path, output_path in zip(file_paths, output_paths):
23
- # 데이터 로드
24
- data = pd.read_csv(file_path, index_col=0)
25
- data= data.loc[data['year'].isin([2018,2019]),:]
26
- data['cloudcover'] = data['cloudcover'].astype('int')
27
- data['lm_cloudcover'] = data['lm_cloudcover'].astype('int')
28
- X = data.drop(columns=['multi_class', 'binary_class'])
29
- y = data['multi_class']
30
-
31
- # 불필요한 열 제거
32
- X.drop(columns=['ground_temp - temp_C', 'hour_sin', 'hour_cos', 'month_sin', 'month_cos'], inplace=True)
33
-
34
- # SMOTENC에서 사용할 범주형 변수 열 번호 설정
35
- categorical_features_indices = [i for i, dtype in enumerate(X.dtypes) if dtype != 'float64']
36
-
37
- # sampling_strategy 설정
38
- count_class_0 = (y == 0).sum()
39
- count_class_1 = (y == 1).sum()
40
- count_class_2 = (y == 2).sum()
41
- sampling_strategy = {
42
- 0: 500 if count_class_0 <= 500 else 1000,
43
- 1: int(np.ceil(count_class_1 / 100) * 100), # 백의 자리로 올림
44
- 2: count_class_2
45
- }
46
-
47
- # SMOTENC 적용
48
- smotenc = SMOTENC(categorical_features=categorical_features_indices, sampling_strategy=sampling_strategy, random_state=42)
49
- X_resampled, y_resampled = smotenc.fit_resample(X, y)
50
-
51
- # Resampled 데이터 생성
52
- lerp_data = X_resampled.copy()
53
- lerp_data['multi_class'] = y_resampled
54
-
55
- # CTGAN에서 사용할 범주형 변수 열 이름 설정
56
- categorical_features = [
57
- col for col, dtype in zip(lerp_data.columns, lerp_data.dtypes) if dtype != 'float64'
58
- ]
59
-
60
- # Optuna 목적 함수 정의
61
- def objective(trial):
62
- # 하이퍼파라미터 탐색 범위 설정
63
- embedding_dim = trial.suggest_int("embedding_dim", 64, 128)
64
- generator_dim = trial.suggest_categorical("generator_dim", [(64, 64), (128, 128)])
65
- discriminator_dim = trial.suggest_categorical("discriminator_dim", [(64, 64), (128, 128)])
66
- pac = trial.suggest_categorical("pac", [4, 8])
67
- batch_size = trial.suggest_categorical("batch_size", [64, 128, 256])
68
- discriminator_steps = trial.suggest_int("discriminator_steps", 1, 3)
69
-
70
- # CTGAN 모델 생성
71
- ctgan = CTGAN(
72
- embedding_dim=embedding_dim,
73
- generator_dim=generator_dim,
74
- discriminator_dim=discriminator_dim,
75
- batch_size=batch_size,
76
- discriminator_steps=discriminator_steps,
77
- pac=pac
78
- )
79
-
80
- # 범주 0 데이터 필터링
81
- data_0 = lerp_data[lerp_data['multi_class'] == 0]
82
-
83
- # 모델 학습
84
- ctgan.fit(data_0, discrete_columns=categorical_features)
85
-
86
- # 샘플 생성
87
- generated_data = ctgan.sample(len(data_0) * 2)
88
-
89
- # 평가: 샘플의 연속형 변수 분포 비교
90
- real_visi = data_0['visi']
91
- generated_visi = generated_data['visi']
92
-
93
- # 분포 간 차이(MSE) 계산
94
- mse = ((real_visi.mean() - generated_visi.mean())**2 + (real_visi.std() - generated_visi.std())**2)
95
- return -mse
96
-
97
- # Optuna로 최적화 수행
98
- study = optuna.create_study(direction="maximize")
99
- study.optimize(objective, n_trials=50)
100
-
101
- # 최적 하이퍼파라미터 출력
102
- best_params = study.best_params
103
-
104
- # 최적 하이퍼파라미터로 CTGAN 학습 및 샘플 생성
105
- ctgan = CTGAN(
106
- embedding_dim=best_params["embedding_dim"],
107
- generator_dim=best_params["generator_dim"],
108
- discriminator_dim=best_params["discriminator_dim"],
109
- batch_size=best_params["batch_size"],
110
- discriminator_steps=best_params["discriminator_steps"],
111
- pac=best_params["pac"]
112
- )
113
-
114
- # 범주 0 데이터로 최종 학습
115
- ctgan.fit(lerp_data[lerp_data['multi_class'] == 0], discrete_columns=categorical_features)
116
- generated_0 = ctgan.sample(7000)
117
-
118
- # 범주 1 데이터 최적화 및 생성
119
- def objective_class1(trial):
120
- embedding_dim = trial.suggest_int("embedding_dim", 128, 512)
121
- generator_dim = trial.suggest_categorical("generator_dim", [(128, 128), (256, 256)])
122
- discriminator_dim = trial.suggest_categorical("discriminator_dim", [(128, 128), (256, 256)])
123
- pac = trial.suggest_categorical("pac", [4, 8])
124
- batch_size = trial.suggest_categorical("batch_size", [256, 512, 1024])
125
- discriminator_steps = trial.suggest_int("discriminator_steps", 1, 5)
126
-
127
- ctgan = CTGAN(
128
- embedding_dim=embedding_dim,
129
- generator_dim=generator_dim,
130
- discriminator_dim=discriminator_dim,
131
- batch_size=batch_size,
132
- discriminator_steps=discriminator_steps,
133
- pac=pac
134
- )
135
-
136
- data_1 = lerp_data[lerp_data['multi_class'] == 1]
137
- ctgan.fit(data_1, discrete_columns=categorical_features)
138
- generated_data = ctgan.sample(len(data_1) * 2)
139
-
140
- real_visi = data_1['visi']
141
- generated_visi = generated_data['visi']
142
- mse = ((real_visi.mean() - generated_visi.mean())**2 + (real_visi.std() - generated_visi.std())**2)
143
- return -mse
144
-
145
- study_class1 = optuna.create_study(direction="maximize")
146
- study_class1.optimize(objective_class1, n_trials=30)
147
-
148
- best_params_class1 = study_class1.best_params
149
- ctgan = CTGAN(
150
- embedding_dim=best_params_class1["embedding_dim"],
151
- generator_dim=best_params_class1["generator_dim"],
152
- discriminator_dim=best_params_class1["discriminator_dim"],
153
- batch_size=best_params_class1["batch_size"],
154
- discriminator_steps=best_params_class1["discriminator_steps"],
155
- pac=best_params_class1["pac"]
156
- )
157
-
158
- ctgan.fit(lerp_data[lerp_data['multi_class'] == 1], discrete_columns=categorical_features)
159
- generated_1 = ctgan.sample(7000 - int(np.ceil(count_class_1 / 100) * 100))
160
-
161
- # 데이터 병합 및 저장
162
- well_generated0 = generated_0[(generated_0['visi'] >= 0) & (generated_0['visi'] < 100)]
163
- well_generated1 = generated_1[(generated_1['visi'] >= 100) & (generated_1['visi'] < 500)]
164
- smote_gan_data = pd.concat([lerp_data, well_generated0, well_generated1], axis=0)
165
- # 제거변수 복구
166
- smote_gan_data['binary_class'] = smote_gan_data['multi_class'].apply(lambda x: 0 if x == 2 else 1)
167
- smote_gan_data['hour_sin'] = np.sin(2 * np.pi * smote_gan_data['hour'] / 24)
168
- smote_gan_data['hour_cos'] = np.cos(2 * np.pi * smote_gan_data['hour'] / 24)
169
- smote_gan_data['month_sin'] = np.sin(2 * np.pi * smote_gan_data['month'] / 12)
170
- smote_gan_data['month_cos'] = np.cos(2 * np.pi * smote_gan_data['month'] / 12)
171
- smote_gan_data['ground_temp - temp_C'] = smote_gan_data['groundtemp'] - smote_gan_data['temp_C']
172
-
173
- filtered_data = smote_gan_data[smote_gan_data['multi_class'] != 2]
174
- original_class2 = data[data['multi_class'] == 2]
175
- final_data = pd.concat([filtered_data, original_class2], axis=0)
176
- final_data.reset_index(drop=True, inplace=True)
177
-
178
- # 결과 저장
179
- final_data.to_csv(output_path, index = False)
180
- print(len(final_data[final_data['multi_class']==0]),'|',len(final_data[final_data['multi_class']==1]),'|',len(final_data[final_data['multi_class']==2]))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/make_oversample_data/gan_sample_7000_2.py DELETED
@@ -1,182 +0,0 @@
1
- import pandas as pd
2
- import numpy as np
3
- import os
4
- os.environ["CUDA_VISIBLE_DEVICES"] = "1"
5
- from imblearn.over_sampling import SMOTENC
6
- import optuna
7
- from ctgan import CTGAN
8
- import torch
9
- import warnings
10
-
11
- # 지역별 데이터 파일 경로
12
- regions = ['incheon', 'seoul','busan', 'daegu', 'daejeon', 'gwangju']
13
- file_paths = [f'../../data/data_for_modeling/{region}_train.csv' for region in regions]
14
- output_paths = [f'../../data/data_oversampled/ctgan7000/ctgan7000_2_{region}.csv' for region in regions]
15
-
16
- # GPU 사용 설정
17
- device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
18
- print(f"Using device: {device}")
19
-
20
- # 경고 무시
21
- warnings.filterwarnings("ignore", category=UserWarning, module="optuna.distributions")
22
-
23
- # 지역별 처리
24
- for file_path, output_path in zip(file_paths, output_paths):
25
- # 데이터 로드
26
- data = pd.read_csv(file_path, index_col=0)
27
- data= data.loc[data['year'].isin([2018,2020]),:]
28
- data['cloudcover'] = data['cloudcover'].astype('int')
29
- data['lm_cloudcover'] = data['lm_cloudcover'].astype('int')
30
- X = data.drop(columns=['multi_class', 'binary_class'])
31
- y = data['multi_class']
32
-
33
- # 불필요한 열 제거
34
- X.drop(columns=['ground_temp - temp_C', 'hour_sin', 'hour_cos', 'month_sin', 'month_cos'], inplace=True)
35
-
36
- # SMOTENC에서 사용할 범주형 변수 열 번호 설정
37
- categorical_features_indices = [i for i, dtype in enumerate(X.dtypes) if dtype != 'float64']
38
-
39
- # sampling_strategy 설정
40
- count_class_0 = (y == 0).sum()
41
- count_class_1 = (y == 1).sum()
42
- count_class_2 = (y == 2).sum()
43
- sampling_strategy = {
44
- 0: 500 if count_class_0 <= 500 else 1000,
45
- 1: int(np.ceil(count_class_1 / 100) * 100), # 백의 자리로 올림
46
- 2: count_class_2
47
- }
48
-
49
- # SMOTENC 적용
50
- smotenc = SMOTENC(categorical_features=categorical_features_indices, sampling_strategy=sampling_strategy, random_state=42)
51
- X_resampled, y_resampled = smotenc.fit_resample(X, y)
52
-
53
- # Resampled 데이터 생성
54
- lerp_data = X_resampled.copy()
55
- lerp_data['multi_class'] = y_resampled
56
-
57
- # CTGAN에서 사용할 범주형 변수 열 이름 설정
58
- categorical_features = [
59
- col for col, dtype in zip(lerp_data.columns, lerp_data.dtypes) if dtype != 'float64'
60
- ]
61
-
62
- # Optuna 목적 함수 정의
63
- def objective(trial):
64
- # 하이퍼파라미터 탐색 범위 설정
65
- embedding_dim = trial.suggest_int("embedding_dim", 64, 128)
66
- generator_dim = trial.suggest_categorical("generator_dim", [(64, 64), (128, 128)])
67
- discriminator_dim = trial.suggest_categorical("discriminator_dim", [(64, 64), (128, 128)])
68
- pac = trial.suggest_categorical("pac", [4, 8])
69
- batch_size = trial.suggest_categorical("batch_size", [64, 128, 256])
70
- discriminator_steps = trial.suggest_int("discriminator_steps", 1, 3)
71
-
72
- # CTGAN 모델 생성
73
- ctgan = CTGAN(
74
- embedding_dim=embedding_dim,
75
- generator_dim=generator_dim,
76
- discriminator_dim=discriminator_dim,
77
- batch_size=batch_size,
78
- discriminator_steps=discriminator_steps,
79
- pac=pac
80
- )
81
-
82
- # 범주 0 데이터 필터링
83
- data_0 = lerp_data[lerp_data['multi_class'] == 0]
84
-
85
- # 모델 학습
86
- ctgan.fit(data_0, discrete_columns=categorical_features)
87
-
88
- # 샘플 생성
89
- generated_data = ctgan.sample(len(data_0) * 2)
90
-
91
- # 평가: 샘플의 연속형 변수 분포 비교
92
- real_visi = data_0['visi']
93
- generated_visi = generated_data['visi']
94
-
95
- # 분포 간 차이(MSE) 계산
96
- mse = ((real_visi.mean() - generated_visi.mean())**2 + (real_visi.std() - generated_visi.std())**2)
97
- return -mse
98
-
99
- # Optuna로 최적화 수행
100
- study = optuna.create_study(direction="maximize")
101
- study.optimize(objective, n_trials=50)
102
-
103
- # 최적 하이퍼파라미터 출력
104
- best_params = study.best_params
105
-
106
- # 최적 하이퍼파라미터로 CTGAN 학습 및 샘플 생성
107
- ctgan = CTGAN(
108
- embedding_dim=best_params["embedding_dim"],
109
- generator_dim=best_params["generator_dim"],
110
- discriminator_dim=best_params["discriminator_dim"],
111
- batch_size=best_params["batch_size"],
112
- discriminator_steps=best_params["discriminator_steps"],
113
- pac=best_params["pac"]
114
- )
115
-
116
- # 범주 0 데이터로 최종 학습
117
- ctgan.fit(lerp_data[lerp_data['multi_class'] == 0], discrete_columns=categorical_features)
118
- generated_0 = ctgan.sample(7000)
119
-
120
- # 범주 1 데이터 최적화 및 생성
121
- def objective_class1(trial):
122
- embedding_dim = trial.suggest_int("embedding_dim", 128, 512)
123
- generator_dim = trial.suggest_categorical("generator_dim", [(128, 128), (256, 256)])
124
- discriminator_dim = trial.suggest_categorical("discriminator_dim", [(128, 128), (256, 256)])
125
- pac = trial.suggest_categorical("pac", [4, 8])
126
- batch_size = trial.suggest_categorical("batch_size", [256, 512, 1024])
127
- discriminator_steps = trial.suggest_int("discriminator_steps", 1, 5)
128
-
129
- ctgan = CTGAN(
130
- embedding_dim=embedding_dim,
131
- generator_dim=generator_dim,
132
- discriminator_dim=discriminator_dim,
133
- batch_size=batch_size,
134
- discriminator_steps=discriminator_steps,
135
- pac=pac
136
- )
137
-
138
- data_1 = lerp_data[lerp_data['multi_class'] == 1]
139
- ctgan.fit(data_1, discrete_columns=categorical_features)
140
- generated_data = ctgan.sample(len(data_1) * 2)
141
-
142
- real_visi = data_1['visi']
143
- generated_visi = generated_data['visi']
144
- mse = ((real_visi.mean() - generated_visi.mean())**2 + (real_visi.std() - generated_visi.std())**2)
145
- return -mse
146
-
147
- study_class1 = optuna.create_study(direction="maximize")
148
- study_class1.optimize(objective_class1, n_trials=30)
149
-
150
- best_params_class1 = study_class1.best_params
151
- ctgan = CTGAN(
152
- embedding_dim=best_params_class1["embedding_dim"],
153
- generator_dim=best_params_class1["generator_dim"],
154
- discriminator_dim=best_params_class1["discriminator_dim"],
155
- batch_size=best_params_class1["batch_size"],
156
- discriminator_steps=best_params_class1["discriminator_steps"],
157
- pac=best_params_class1["pac"]
158
- )
159
-
160
- ctgan.fit(lerp_data[lerp_data['multi_class'] == 1], discrete_columns=categorical_features)
161
- generated_1 = ctgan.sample(7000 - int(np.ceil(count_class_1 / 100) * 100))
162
-
163
- # 데이터 병합 및 저장
164
- well_generated0 = generated_0[(generated_0['visi'] >= 0) & (generated_0['visi'] < 100)]
165
- well_generated1 = generated_1[(generated_1['visi'] >= 100) & (generated_1['visi'] < 500)]
166
- smote_gan_data = pd.concat([lerp_data, well_generated0, well_generated1], axis=0)
167
- # 제거변수 복구
168
- smote_gan_data['binary_class'] = smote_gan_data['multi_class'].apply(lambda x: 0 if x == 2 else 1)
169
- smote_gan_data['hour_sin'] = np.sin(2 * np.pi * smote_gan_data['hour'] / 24)
170
- smote_gan_data['hour_cos'] = np.cos(2 * np.pi * smote_gan_data['hour'] / 24)
171
- smote_gan_data['month_sin'] = np.sin(2 * np.pi * smote_gan_data['month'] / 12)
172
- smote_gan_data['month_cos'] = np.cos(2 * np.pi * smote_gan_data['month'] / 12)
173
- smote_gan_data['ground_temp - temp_C'] = smote_gan_data['groundtemp'] - smote_gan_data['temp_C']
174
-
175
- filtered_data = smote_gan_data[smote_gan_data['multi_class'] != 2]
176
- original_class2 = data[data['multi_class'] == 2]
177
- final_data = pd.concat([filtered_data, original_class2], axis=0)
178
- final_data.reset_index(drop=True, inplace=True)
179
-
180
- # 결과 저장
181
- final_data.to_csv(output_path, index = False)
182
- print(len(final_data[final_data['multi_class']==0]),'|',len(final_data[final_data['multi_class']==1]),'|',len(final_data[final_data['multi_class']==2]))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/make_oversample_data/gan_sample_7000_3.py DELETED
@@ -1,182 +0,0 @@
1
- import pandas as pd
2
- import numpy as np
3
- import os
4
- os.environ["CUDA_VISIBLE_DEVICES"] = "0"
5
- from imblearn.over_sampling import SMOTENC
6
- import optuna
7
- from ctgan import CTGAN
8
- import torch
9
- import warnings
10
-
11
- # 지역별 데이터 파일 경로
12
- regions = ['incheon', 'seoul','busan', 'daegu', 'daejeon', 'gwangju']
13
- file_paths = [f'../../data/data_for_modeling/{region}_train.csv' for region in regions]
14
- output_paths = [f'../../data/data_oversampled/ctgan7000/ctgan7000_3_{region}.csv' for region in regions]
15
-
16
- # GPU 사용 설정
17
- device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
18
- print(f"Using device: {device}")
19
-
20
- # 경고 무시
21
- warnings.filterwarnings("ignore", category=UserWarning, module="optuna.distributions")
22
-
23
- # 지역별 처리
24
- for file_path, output_path in zip(file_paths, output_paths):
25
- # 데이터 로드
26
- data = pd.read_csv(file_path, index_col=0)
27
- data= data.loc[data['year'].isin([2019,2020]),:]
28
- data['cloudcover'] = data['cloudcover'].astype('int')
29
- data['lm_cloudcover'] = data['lm_cloudcover'].astype('int')
30
- X = data.drop(columns=['multi_class', 'binary_class'])
31
- y = data['multi_class']
32
-
33
- # 불필요한 열 제거
34
- X.drop(columns=['ground_temp - temp_C', 'hour_sin', 'hour_cos', 'month_sin', 'month_cos'], inplace=True)
35
-
36
- # SMOTENC에서 사용할 범주형 변수 열 번호 설정
37
- categorical_features_indices = [i for i, dtype in enumerate(X.dtypes) if dtype != 'float64']
38
-
39
- # sampling_strategy 설정
40
- count_class_0 = (y == 0).sum()
41
- count_class_1 = (y == 1).sum()
42
- count_class_2 = (y == 2).sum()
43
- sampling_strategy = {
44
- 0: 500 if count_class_0 <= 500 else 1000,
45
- 1: int(np.ceil(count_class_1 / 100) * 100), # 백의 자리로 올림
46
- 2: count_class_2
47
- }
48
-
49
- # SMOTENC 적용
50
- smotenc = SMOTENC(categorical_features=categorical_features_indices, sampling_strategy=sampling_strategy, random_state=42)
51
- X_resampled, y_resampled = smotenc.fit_resample(X, y)
52
-
53
- # Resampled 데이터 생성
54
- lerp_data = X_resampled.copy()
55
- lerp_data['multi_class'] = y_resampled
56
-
57
- # CTGAN에서 사용할 범주형 변수 열 이름 설정
58
- categorical_features = [
59
- col for col, dtype in zip(lerp_data.columns, lerp_data.dtypes) if dtype != 'float64'
60
- ]
61
-
62
- # Optuna 목적 함수 정의
63
- def objective(trial):
64
- # 하이퍼파라미터 탐색 범위 설정
65
- embedding_dim = trial.suggest_int("embedding_dim", 64, 128)
66
- generator_dim = trial.suggest_categorical("generator_dim", [(64, 64), (128, 128)])
67
- discriminator_dim = trial.suggest_categorical("discriminator_dim", [(64, 64), (128, 128)])
68
- pac = trial.suggest_categorical("pac", [4, 8])
69
- batch_size = trial.suggest_categorical("batch_size", [64, 128, 256])
70
- discriminator_steps = trial.suggest_int("discriminator_steps", 1, 3)
71
-
72
- # CTGAN 모델 생성
73
- ctgan = CTGAN(
74
- embedding_dim=embedding_dim,
75
- generator_dim=generator_dim,
76
- discriminator_dim=discriminator_dim,
77
- batch_size=batch_size,
78
- discriminator_steps=discriminator_steps,
79
- pac=pac
80
- )
81
-
82
- # 범주 0 데이터 필터링
83
- data_0 = lerp_data[lerp_data['multi_class'] == 0]
84
-
85
- # 모델 학습
86
- ctgan.fit(data_0, discrete_columns=categorical_features)
87
-
88
- # 샘플 생성
89
- generated_data = ctgan.sample(len(data_0) * 2)
90
-
91
- # 평가: 샘플의 연속형 변수 분포 비교
92
- real_visi = data_0['visi']
93
- generated_visi = generated_data['visi']
94
-
95
- # 분포 간 차이(MSE) 계산
96
- mse = ((real_visi.mean() - generated_visi.mean())**2 + (real_visi.std() - generated_visi.std())**2)
97
- return -mse
98
-
99
- # Optuna로 최적화 수행
100
- study = optuna.create_study(direction="maximize")
101
- study.optimize(objective, n_trials=50)
102
-
103
- # 최적 하이퍼파라미터 출력
104
- best_params = study.best_params
105
-
106
- # 최적 하이퍼파라미터로 CTGAN 학습 및 샘플 생성
107
- ctgan = CTGAN(
108
- embedding_dim=best_params["embedding_dim"],
109
- generator_dim=best_params["generator_dim"],
110
- discriminator_dim=best_params["discriminator_dim"],
111
- batch_size=best_params["batch_size"],
112
- discriminator_steps=best_params["discriminator_steps"],
113
- pac=best_params["pac"]
114
- )
115
-
116
- # 범주 0 데이터로 최종 학습
117
- ctgan.fit(lerp_data[lerp_data['multi_class'] == 0], discrete_columns=categorical_features)
118
- generated_0 = ctgan.sample(7000)
119
-
120
- # 범주 1 데이터 최적화 및 생성
121
- def objective_class1(trial):
122
- embedding_dim = trial.suggest_int("embedding_dim", 128, 512)
123
- generator_dim = trial.suggest_categorical("generator_dim", [(128, 128), (256, 256)])
124
- discriminator_dim = trial.suggest_categorical("discriminator_dim", [(128, 128), (256, 256)])
125
- pac = trial.suggest_categorical("pac", [4, 8])
126
- batch_size = trial.suggest_categorical("batch_size", [256, 512, 1024])
127
- discriminator_steps = trial.suggest_int("discriminator_steps", 1, 5)
128
-
129
- ctgan = CTGAN(
130
- embedding_dim=embedding_dim,
131
- generator_dim=generator_dim,
132
- discriminator_dim=discriminator_dim,
133
- batch_size=batch_size,
134
- discriminator_steps=discriminator_steps,
135
- pac=pac
136
- )
137
-
138
- data_1 = lerp_data[lerp_data['multi_class'] == 1]
139
- ctgan.fit(data_1, discrete_columns=categorical_features)
140
- generated_data = ctgan.sample(len(data_1) * 2)
141
-
142
- real_visi = data_1['visi']
143
- generated_visi = generated_data['visi']
144
- mse = ((real_visi.mean() - generated_visi.mean())**2 + (real_visi.std() - generated_visi.std())**2)
145
- return -mse
146
-
147
- study_class1 = optuna.create_study(direction="maximize")
148
- study_class1.optimize(objective_class1, n_trials=30)
149
-
150
- best_params_class1 = study_class1.best_params
151
- ctgan = CTGAN(
152
- embedding_dim=best_params_class1["embedding_dim"],
153
- generator_dim=best_params_class1["generator_dim"],
154
- discriminator_dim=best_params_class1["discriminator_dim"],
155
- batch_size=best_params_class1["batch_size"],
156
- discriminator_steps=best_params_class1["discriminator_steps"],
157
- pac=best_params_class1["pac"]
158
- )
159
-
160
- ctgan.fit(lerp_data[lerp_data['multi_class'] == 1], discrete_columns=categorical_features)
161
- generated_1 = ctgan.sample(7000 - int(np.ceil(count_class_1 / 100) * 100))
162
-
163
- # 데이터 병합 및 저장
164
- well_generated0 = generated_0[(generated_0['visi'] >= 0) & (generated_0['visi'] < 100)]
165
- well_generated1 = generated_1[(generated_1['visi'] >= 100) & (generated_1['visi'] < 500)]
166
- smote_gan_data = pd.concat([lerp_data, well_generated0, well_generated1], axis=0)
167
- # 제거변수 복구
168
- smote_gan_data['binary_class'] = smote_gan_data['multi_class'].apply(lambda x: 0 if x == 2 else 1)
169
- smote_gan_data['hour_sin'] = np.sin(2 * np.pi * smote_gan_data['hour'] / 24)
170
- smote_gan_data['hour_cos'] = np.cos(2 * np.pi * smote_gan_data['hour'] / 24)
171
- smote_gan_data['month_sin'] = np.sin(2 * np.pi * smote_gan_data['month'] / 12)
172
- smote_gan_data['month_cos'] = np.cos(2 * np.pi * smote_gan_data['month'] / 12)
173
- smote_gan_data['ground_temp - temp_C'] = smote_gan_data['groundtemp'] - smote_gan_data['temp_C']
174
-
175
- filtered_data = smote_gan_data[smote_gan_data['multi_class'] != 2]
176
- original_class2 = data[data['multi_class'] == 2]
177
- final_data = pd.concat([filtered_data, original_class2], axis=0)
178
- final_data.reset_index(drop=True, inplace=True)
179
-
180
- # 결과 저장
181
- final_data.to_csv(output_path, index = False)
182
- print(len(final_data[final_data['multi_class']==0]),'|',len(final_data[final_data['multi_class']==1]),'|',len(final_data[final_data['multi_class']==2]))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/make_oversample_data/oversampling_code.py DELETED
@@ -1,355 +0,0 @@
1
- import pandas as pd
2
- import numpy as np
3
- from imblearn.over_sampling import SMOTENC
4
-
5
-
6
- # smote와 ctgan을 이용한 oversampling 진행
7
-
8
- # 파일 경로와 지역 이름 리스트
9
- regions = ['busan', 'daegu', 'daejeon', 'incheon', 'seoul','gwangju']
10
- input_paths = [f'../data/data_for_modeling/{region}_train.csv' for region in regions]
11
-
12
-
13
-
14
-
15
- # 반복적으로 각 지역 데이터 처리
16
- for region, input_path in zip(regions, input_paths):
17
- # 데이터 읽기
18
- data = pd.read_csv(input_path, index_col=0)
19
- data.drop(['Unnamed: 0'], axis=1, inplace=True)
20
- print("\n######",region,"#######")
21
- print(len(data[data['multi_class']==0]),'|',len(data[data['multi_class']==1]),'|',len(data[data['multi_class']==2]))
22
- print(len(data.columns))
23
-
24
-
25
-
26
-
27
-
28
- import pandas as pd
29
- import numpy as np
30
- from imblearn.over_sampling import SMOTENC
31
-
32
- # 파일 경로와 지역 이름 리스트
33
- regions = ['busan', 'daegu', 'daejeon', 'incheon', 'seoul','gwangju']
34
- input_paths = [f'../data/data_for_modeling/{region}_train.csv' for region in regions]
35
- output_paths = [f'../data/data_oversampled/smote_{region}.csv' for region in regions]
36
-
37
- # 반복적으로 각 지역 데이터 처리
38
- for region, input_path, output_path in zip(regions, input_paths, output_paths):
39
- # 데이터 읽기
40
- data = pd.read_csv(input_path, index_col=0)
41
- data.drop(['Unnamed: 0'], axis=1, inplace=True)
42
-
43
- # X와 y 분리
44
- X = data.drop(columns=['multi_class', 'binary_class'])
45
- y = data['multi_class']
46
-
47
- # 불필요한 열 제거
48
- X.drop(columns=['ground_temp - temp_C', 'hour_sin', 'hour_cos', 'month_sin', 'month_cos'], inplace=True)
49
-
50
- # 범주형 변수 식별
51
- categorical_features = [i for i, dtype in enumerate(X.dtypes) if dtype != 'float64']
52
-
53
- # 각 지역의 multi_class 값이 2인 데이터 개수 계산
54
- count_class_2 = (y == 2).sum()
55
-
56
- # SMOTENC 적용
57
- smotenc = SMOTENC(
58
- categorical_features=categorical_features,
59
- sampling_strategy={0: 10000, 1: 10000, 2: count_class_2},
60
- random_state=42
61
- )
62
- X_resampled, y_resampled = smotenc.fit_resample(X, y)
63
-
64
- # 추가 변수 생성
65
- X_resampled['multi_class'] = y_resampled
66
- X_resampled['binary_class'] = X_resampled['multi_class'].apply(lambda x: 0 if x == 2 else 1)
67
- X_resampled['hour_sin'] = np.sin(2 * np.pi * X_resampled['hour'] / 24)
68
- X_resampled['hour_cos'] = np.cos(2 * np.pi * X_resampled['hour'] / 24)
69
- X_resampled['month_sin'] = np.sin(2 * np.pi * X_resampled['month'] / 12)
70
- X_resampled['month_cos'] = np.cos(2 * np.pi * X_resampled['month'] / 12)
71
- X_resampled['ground_temp - temp_C'] = X_resampled['groundtemp'] - X_resampled['temp_C']
72
-
73
- # 결과 저장
74
- X_resampled.to_csv(output_path)
75
- print(f"Processed and saved: {region} -> {output_path}")
76
-
77
-
78
- smote_seoul = pd.read_csv('../data/data_oversampled/smote_seoul.csv')
79
- print(smote_seoul[smote_seoul['multi_class']==0]['visi'].describe())
80
- print(smote_seoul[smote_seoul['multi_class']==1]['visi'].describe())
81
-
82
-
83
- import pandas as pd
84
- import numpy as np
85
- from imblearn.over_sampling import SMOTENC
86
-
87
- # 파일 경로와 지역 이름 리스트
88
- regions = ['busan', 'daegu', 'daejeon', 'incheon', 'seoul','gwangju']
89
- input_paths = [f'../data/data_oversampled/smote_{region}.csv' for region in regions]
90
-
91
- # 반복적으로 각 지역 데이터 처리
92
- for region, input_path in zip(regions, input_paths):
93
- # 데이터 읽기
94
- data = pd.read_csv(input_path, index_col=0)
95
- data.drop(['Unnamed: 0'], axis=1, inplace=True)
96
- print("\n######",region,"#######")
97
- print(len(data[data['multi_class']==0]),'|',len(data[data['multi_class']==1]),'|',len(data[data['multi_class']==2]))
98
- print(len(data.columns))
99
-
100
-
101
-
102
- import pandas as pd
103
- import numpy as np
104
- from imblearn.over_sampling import SMOTENC
105
-
106
- # 파일 경로와 지역 이름 리스트
107
- regions = ['busan', 'daegu', 'daejeon', 'incheon', 'seoul','gwangju']
108
- input_paths = [f'../data/data_for_modeling/{region}_train.csv' for region in regions]
109
-
110
- # 반복적으로 각 지역 데이터 처리
111
- for region, input_path in zip(regions, input_paths):
112
- # 데이터 읽기
113
- data = pd.read_csv(input_path, index_col=0)
114
- data.drop(['Unnamed: 0'], axis=1, inplace=True)
115
- print("\n######",region,"#######")
116
- print(len(data[data['multi_class']==0]),'|',len(data[data['multi_class']==1]),'|',len(data[data['multi_class']==2]))
117
- print(len(data.columns))
118
-
119
-
120
- import pandas as pd
121
- import numpy as np
122
- from imblearn.over_sampling import SMOTENC
123
- import optuna
124
- from ctgan import CTGAN
125
- import torch
126
- import warnings
127
-
128
- # 지역별 데이터 파일 경로
129
- regions = ['busan', 'daegu', 'daejeon', 'incheon', 'seoul','gwangju']
130
- file_paths = [f'../data/data_for_modeling/df_{region}.feather' for region in regions]
131
- output_paths = [f'../data/data_oversampled/ctgan_{region}.csv' for region in regions]
132
-
133
- # GPU 사용 설정
134
- device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
135
- print(f"Using device: {device}")
136
-
137
- # 경고 무시
138
- warnings.filterwarnings("ignore", category=UserWarning, module="optuna.distributions")
139
-
140
- # 지역별 처리
141
- for file_path, output_path in zip(file_paths, output_paths):
142
- # 데이터 로드
143
- data = pd.read_feather(file_path)
144
- data.drop(['Unnamed: 0'], axis=1, inplace=True)
145
- X = data.drop(columns=['multi_class', 'binary_class'])
146
- y = data['multi_class']
147
-
148
- # 불필요한 열 제거
149
- X.drop(columns=['ground_temp - temp_C', 'hour_sin', 'hour_cos', 'month_sin', 'month_cos'], inplace=True)
150
-
151
- # SMOTENC에서 사용할 범주형 변수 열 번호 설정
152
- categorical_features_indices = [i for i, dtype in enumerate(X.dtypes) if dtype != 'float64']
153
-
154
- # sampling_strategy 설정
155
- count_class_0 = (y == 0).sum()
156
- count_class_1 = (y == 1).sum()
157
- count_class_2 = (y == 2).sum()
158
- sampling_strategy = {
159
- 0: 500 if count_class_0 <= 500 else 1000,
160
- 1: int(np.ceil(count_class_1 / 100) * 100), # 백의 자리로 올림
161
- 2: count_class_2
162
- }
163
-
164
- # SMOTENC 적용
165
- smotenc = SMOTENC(categorical_features=categorical_features_indices, sampling_strategy=sampling_strategy, random_state=42)
166
- X_resampled, y_resampled = smotenc.fit_resample(X, y)
167
-
168
- # Resampled 데이터 생성
169
- lerp_data = X_resampled.copy()
170
- lerp_data['multi_class'] = y_resampled
171
-
172
- # CTGAN에서 사용할 범주형 변수 열 이름 설정
173
- categorical_features = [
174
- col for col, dtype in zip(lerp_data.columns, lerp_data.dtypes) if dtype != 'float64'
175
- ]
176
-
177
- # Optuna 목적 함수 정의
178
- def objective(trial):
179
- # 하이퍼파라미터 탐색 범위 설정
180
- embedding_dim = trial.suggest_int("embedding_dim", 64, 128)
181
- generator_dim = trial.suggest_categorical("generator_dim", [(64, 64), (128, 128)])
182
- discriminator_dim = trial.suggest_categorical("discriminator_dim", [(64, 64), (128, 128)])
183
- pac = trial.suggest_categorical("pac", [4, 8])
184
- batch_size = trial.suggest_categorical("batch_size", [64, 128, 256])
185
- discriminator_steps = trial.suggest_int("discriminator_steps", 1, 3)
186
-
187
- # CTGAN 모델 생성
188
- ctgan = CTGAN(
189
- embedding_dim=embedding_dim,
190
- generator_dim=generator_dim,
191
- discriminator_dim=discriminator_dim,
192
- batch_size=batch_size,
193
- discriminator_steps=discriminator_steps,
194
- pac=pac
195
- )
196
-
197
- # 범주 0 데이터 필터링
198
- data_0 = lerp_data[lerp_data['multi_class'] == 0]
199
-
200
- # 모델 학습
201
- ctgan.fit(data_0, discrete_columns=categorical_features)
202
-
203
- # 샘플 생성
204
- generated_data = ctgan.sample(len(data_0) * 2)
205
-
206
- # 평가: 샘플의 연속형 변수 분포 비교
207
- real_visi = data_0['visi']
208
- generated_visi = generated_data['visi']
209
-
210
- # 분포 간 차이(MSE) 계산
211
- mse = ((real_visi.mean() - generated_visi.mean())**2 + (real_visi.std() - generated_visi.std())**2)
212
- return -mse
213
-
214
- # Optuna로 최적화 수행
215
- study = optuna.create_study(direction="maximize")
216
- study.optimize(objective, n_trials=50)
217
-
218
- # 최적 하이퍼파라미터 출력
219
- best_params = study.best_params
220
-
221
- # 최적 하이퍼파라미터로 CTGAN 학습 및 샘플 생성
222
- ctgan = CTGAN(
223
- embedding_dim=best_params["embedding_dim"],
224
- generator_dim=best_params["generator_dim"],
225
- discriminator_dim=best_params["discriminator_dim"],
226
- batch_size=best_params["batch_size"],
227
- discriminator_steps=best_params["discriminator_steps"],
228
- pac=best_params["pac"]
229
- )
230
-
231
- # 범주 0 데이터로 최종 학습
232
- ctgan.fit(lerp_data[lerp_data['multi_class'] == 0], discrete_columns=categorical_features)
233
- generated_0 = ctgan.sample(19500 if count_class_0 <= 500 else 19000)
234
-
235
- # 범주 1 데이터 최적화 및 생성
236
- def objective_class1(trial):
237
- embedding_dim = trial.suggest_int("embedding_dim", 128, 512)
238
- generator_dim = trial.suggest_categorical("generator_dim", [(128, 128), (256, 256)])
239
- discriminator_dim = trial.suggest_categorical("discriminator_dim", [(128, 128), (256, 256)])
240
- pac = trial.suggest_categorical("pac", [4, 8])
241
- batch_size = trial.suggest_categorical("batch_size", [256, 512, 1024])
242
- discriminator_steps = trial.suggest_int("discriminator_steps", 1, 5)
243
-
244
- ctgan = CTGAN(
245
- embedding_dim=embedding_dim,
246
- generator_dim=generator_dim,
247
- discriminator_dim=discriminator_dim,
248
- batch_size=batch_size,
249
- discriminator_steps=discriminator_steps,
250
- pac=pac
251
- )
252
-
253
- data_1 = lerp_data[lerp_data['multi_class'] == 1]
254
- ctgan.fit(data_1, discrete_columns=categorical_features)
255
- generated_data = ctgan.sample(len(data_1) * 2)
256
-
257
- real_visi = data_1['visi']
258
- generated_visi = generated_data['visi']
259
- mse = ((real_visi.mean() - generated_visi.mean())**2 + (real_visi.std() - generated_visi.std())**2)
260
- return -mse
261
-
262
- study_class1 = optuna.create_study(direction="maximize")
263
- study_class1.optimize(objective_class1, n_trials=30)
264
-
265
- best_params_class1 = study_class1.best_params
266
- ctgan = CTGAN(
267
- embedding_dim=best_params_class1["embedding_dim"],
268
- generator_dim=best_params_class1["generator_dim"],
269
- discriminator_dim=best_params_class1["discriminator_dim"],
270
- batch_size=best_params_class1["batch_size"],
271
- discriminator_steps=best_params_class1["discriminator_steps"],
272
- pac=best_params_class1["pac"]
273
- )
274
-
275
- ctgan.fit(lerp_data[lerp_data['multi_class'] == 1], discrete_columns=categorical_features)
276
- generated_1 = ctgan.sample(20000 - int(np.ceil(count_class_1 / 100) * 100))
277
-
278
- # 데이터 병합 및 저장
279
- well_generated0 = generated_0[(generated_0['visi'] >= 0) & (generated_0['visi'] < 100)]
280
- well_generated1 = generated_1[(generated_1['visi'] >= 100) & (generated_1['visi'] < 500)]
281
- smote_gan_data = pd.concat([lerp_data, well_generated0, well_generated1], axis=0)
282
- # 제거변수 복구
283
- smote_gan_data['binary_class'] = smote_gan_data['multi_class'].apply(lambda x: 0 if x == 2 else 1)
284
- smote_gan_data['hour_sin'] = np.sin(2 * np.pi * smote_gan_data['hour'] / 24)
285
- smote_gan_data['hour_cos'] = np.cos(2 * np.pi * smote_gan_data['hour'] / 24)
286
- smote_gan_data['month_sin'] = np.sin(2 * np.pi * smote_gan_data['month'] / 12)
287
- smote_gan_data['month_cos'] = np.cos(2 * np.pi * smote_gan_data['month'] / 12)
288
- smote_gan_data['ground_temp - temp_C'] = smote_gan_data['groundtemp'] - smote_gan_data['temp_C']
289
-
290
- # 결과 저장
291
- smote_gan_data.to_csv(output_path, index = False)
292
- print(f"Processed and saved: {region} -> {output_path}")
293
-
294
-
295
-
296
- import pandas as pd
297
- import numpy as np
298
- from imblearn.over_sampling import SMOTENC
299
-
300
- # 파일 경로와 지역 이름 리스트
301
- regions = ['busan', 'daegu', 'daejeon', 'incheon', 'seoul','gwangju']
302
- input_paths = [f'../data/data_oversampled/ctgan_{region}.csv' for region in regions]
303
-
304
- # 반복적으로 각 지역 데이터 처리
305
- for region, input_path in zip(regions, input_paths):
306
- # 데이터 읽기
307
- data = pd.read_csv(input_path)
308
- print("\n######",region,"#######")
309
- print(len(data[data['multi_class']==0]),'|',len(data[data['multi_class']==1]),'|',len(data[data['multi_class']==2]))
310
- print(len(data.columns))
311
-
312
-
313
-
314
-
315
- busan_check = pd.read_csv('../data/data_oversampled/ctgan_busan.csv')
316
- print(busan_check[busan_check['multi_class']==0]['visi'].describe())
317
- print(busan_check[busan_check['multi_class']==1]['visi'].describe())
318
- print(busan_check[busan_check['multi_class']==2]['visi'].describe())
319
-
320
-
321
-
322
-
323
- import pandas as pd
324
- import numpy as np
325
- from imblearn.over_sampling import SMOTENC
326
-
327
- # 파일 경로와 지역 이름 리스트
328
- regions = ['busan', 'daegu', 'daejeon', 'incheon', 'seoul','gwangju']
329
- origin_paths = [f'../data/data_for_modeling/{region}_train.csv' for region in regions]
330
- augment_paths = [f'../data/data_oversampled/ctgan_{region}.csv' for region in regions]
331
-
332
- # 반복적으로 각 지역 데이터 처리
333
- for region, origin_path, augment_path in zip(regions, origin_paths, augment_paths):
334
- # 데이터 읽기
335
- origin = pd.read_csv(origin_path, index_col=0)
336
- augment = pd.read_csv(augment_path)
337
-
338
- # 증강된 데이터에서 범주 2 데이터 제거
339
- filtered_data = augment[augment['multi_class'] != 2]
340
-
341
- # 원본 데이터에서 범주 2 데이터 추출
342
- original_class2 = origin[origin['multi_class'] == 2]
343
-
344
- # 제거된 데이터에 원본 범주 2 데이터를 추가
345
- final_data = pd.concat([filtered_data, original_class2], axis=0)
346
-
347
- # 인덱스 재설정
348
- final_data.reset_index(drop=True, inplace=True)
349
-
350
- # 결과 저장
351
- final_data.to_csv(augment_path, index = False)
352
-
353
- print("\n######",region,"#######")
354
- print(len(final_data[final_data['multi_class']==0]),'|',len(final_data[final_data['multi_class']==1]),'|',len(final_data[final_data['multi_class']==2]))
355
- print(len(data.columns))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/make_oversample_data/smote_sample_1.py DELETED
@@ -1,53 +0,0 @@
1
- import pandas as pd
2
- import numpy as np
3
-
4
- from imblearn.over_sampling import SMOTENC
5
-
6
- # 지역별 데이터 파일 경로
7
- regions = ['incheon', 'seoul','busan', 'daegu', 'daejeon', 'gwangju']
8
- file_paths = [f'../../data/data_for_modeling/{region}_train.csv' for region in regions]
9
- output_paths = [f'../../data/data_oversampled/smote/smote_1_{region}.csv' for region in regions]
10
-
11
- # 지역별 처리
12
- for file_path, output_path in zip(file_paths, output_paths):
13
- # 데이터 로드
14
- data = pd.read_csv(file_path, index_col=0)
15
- data= data.loc[data['year'].isin([2018,2019]),:]
16
- data['cloudcover'] = data['cloudcover'].astype('int')
17
- data['lm_cloudcover'] = data['lm_cloudcover'].astype('int')
18
- X = data.drop(columns=['multi_class', 'binary_class'])
19
- y = data['multi_class']
20
-
21
- # 불필요한 열 제거
22
- X.drop(columns=['ground_temp - temp_C', 'hour_sin', 'hour_cos', 'month_sin', 'month_cos'], inplace=True)
23
-
24
- # SMOTENC에서 사용할 범주형 변수 열 번호 설정
25
- categorical_features_indices = [i for i, dtype in enumerate(X.dtypes) if dtype != 'float64']
26
-
27
- # sampling_strategy 설정
28
- count_class_2 = (y == 2).sum()
29
- sampling_strategy = {
30
- 0: int(np.ceil(count_class_2 / 1000) * 500),
31
- 1: int(np.ceil(count_class_2 / 1000) * 500),
32
- 2: count_class_2
33
- }
34
-
35
- # SMOTENC 적용
36
- smotenc = SMOTENC(categorical_features=categorical_features_indices, sampling_strategy=sampling_strategy, random_state=42)
37
- X_resampled, y_resampled = smotenc.fit_resample(X, y)
38
-
39
- # Resampled 데이터 생성
40
- lerp_data = X_resampled.copy()
41
- lerp_data['multi_class'] = y_resampled
42
-
43
- # 제거변수 복구
44
- lerp_data['binary_class'] = lerp_data['multi_class'].apply(lambda x: 0 if x == 2 else 1)
45
- lerp_data['hour_sin'] = np.sin(2 * np.pi * lerp_data['hour'] / 24)
46
- lerp_data['hour_cos'] = np.cos(2 * np.pi * lerp_data['hour'] / 24)
47
- lerp_data['month_sin'] = np.sin(2 * np.pi * lerp_data['month'] / 12)
48
- lerp_data['month_cos'] = np.cos(2 * np.pi * lerp_data['month'] / 12)
49
- lerp_data['ground_temp - temp_C'] = lerp_data['groundtemp'] - lerp_data['temp_C']
50
-
51
- # 결과 저장
52
- lerp_data.to_csv(output_path, index = False)
53
- print(len(lerp_data[lerp_data['multi_class']==0]),'|',len(lerp_data[lerp_data['multi_class']==1]),'|',len(lerp_data[lerp_data['multi_class']==2]))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/make_oversample_data/smote_sample_2.py DELETED
@@ -1,53 +0,0 @@
1
- import pandas as pd
2
- import numpy as np
3
-
4
- from imblearn.over_sampling import SMOTENC
5
-
6
- # 지역별 데이터 파일 경로
7
- regions = ['incheon', 'seoul','busan', 'daegu', 'daejeon', 'gwangju']
8
- file_paths = [f'../../data/data_for_modeling/{region}_train.csv' for region in regions]
9
- output_paths = [f'../../data/data_oversampled/smote/smote_2_{region}.csv' for region in regions]
10
-
11
- # 지역별 처리
12
- for file_path, output_path in zip(file_paths, output_paths):
13
- # 데이터 로드
14
- data = pd.read_csv(file_path, index_col=0)
15
- data= data.loc[data['year'].isin([2018,2020]),:]
16
- data['cloudcover'] = data['cloudcover'].astype('int')
17
- data['lm_cloudcover'] = data['lm_cloudcover'].astype('int')
18
- X = data.drop(columns=['multi_class', 'binary_class'])
19
- y = data['multi_class']
20
-
21
- # 불필요한 열 제거
22
- X.drop(columns=['ground_temp - temp_C', 'hour_sin', 'hour_cos', 'month_sin', 'month_cos'], inplace=True)
23
-
24
- # SMOTENC에서 사용할 범주형 변수 열 번호 설정
25
- categorical_features_indices = [i for i, dtype in enumerate(X.dtypes) if dtype != 'float64']
26
-
27
- # sampling_strategy 설정
28
- count_class_2 = (y == 2).sum()
29
- sampling_strategy = {
30
- 0: int(np.ceil(count_class_2 / 1000) * 500),
31
- 1: int(np.ceil(count_class_2 / 1000) * 500),
32
- 2: count_class_2
33
- }
34
-
35
- # SMOTENC 적용
36
- smotenc = SMOTENC(categorical_features=categorical_features_indices, sampling_strategy=sampling_strategy, random_state=42)
37
- X_resampled, y_resampled = smotenc.fit_resample(X, y)
38
-
39
- # Resampled 데이터 생성
40
- lerp_data = X_resampled.copy()
41
- lerp_data['multi_class'] = y_resampled
42
-
43
- # 제거변수 복구
44
- lerp_data['binary_class'] = lerp_data['multi_class'].apply(lambda x: 0 if x == 2 else 1)
45
- lerp_data['hour_sin'] = np.sin(2 * np.pi * lerp_data['hour'] / 24)
46
- lerp_data['hour_cos'] = np.cos(2 * np.pi * lerp_data['hour'] / 24)
47
- lerp_data['month_sin'] = np.sin(2 * np.pi * lerp_data['month'] / 12)
48
- lerp_data['month_cos'] = np.cos(2 * np.pi * lerp_data['month'] / 12)
49
- lerp_data['ground_temp - temp_C'] = lerp_data['groundtemp'] - lerp_data['temp_C']
50
-
51
- # 결과 저장
52
- lerp_data.to_csv(output_path, index = False)
53
- print(len(lerp_data[lerp_data['multi_class']==0]),'|',len(lerp_data[lerp_data['multi_class']==1]),'|',len(lerp_data[lerp_data['multi_class']==2]))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/make_oversample_data/smote_sample_3.py DELETED
@@ -1,53 +0,0 @@
1
- import pandas as pd
2
- import numpy as np
3
-
4
- from imblearn.over_sampling import SMOTENC
5
-
6
- # 지역별 데이터 파일 경로
7
- regions = ['incheon', 'seoul','busan', 'daegu', 'daejeon', 'gwangju']
8
- file_paths = [f'../../data/data_for_modeling/{region}_train.csv' for region in regions]
9
- output_paths = [f'../../data/data_oversampled/smote/smote_3_{region}.csv' for region in regions]
10
-
11
- # 지역별 처리
12
- for file_path, output_path in zip(file_paths, output_paths):
13
- # 데이터 로드
14
- data = pd.read_csv(file_path, index_col=0)
15
- data= data.loc[data['year'].isin([2019,2020]),:]
16
- data['cloudcover'] = data['cloudcover'].astype('int')
17
- data['lm_cloudcover'] = data['lm_cloudcover'].astype('int')
18
- X = data.drop(columns=['multi_class', 'binary_class'])
19
- y = data['multi_class']
20
-
21
- # 불필요한 열 제거
22
- X.drop(columns=['ground_temp - temp_C', 'hour_sin', 'hour_cos', 'month_sin', 'month_cos'], inplace=True)
23
-
24
- # SMOTENC에서 사용할 범주형 변수 열 번호 설정
25
- categorical_features_indices = [i for i, dtype in enumerate(X.dtypes) if dtype != 'float64']
26
-
27
- # sampling_strategy 설정
28
- count_class_2 = (y == 2).sum()
29
- sampling_strategy = {
30
- 0: int(np.ceil(count_class_2 / 1000) * 500),
31
- 1: int(np.ceil(count_class_2 / 1000) * 500),
32
- 2: count_class_2
33
- }
34
-
35
- # SMOTENC 적용
36
- smotenc = SMOTENC(categorical_features=categorical_features_indices, sampling_strategy=sampling_strategy, random_state=42)
37
- X_resampled, y_resampled = smotenc.fit_resample(X, y)
38
-
39
- # Resampled 데이터 생성
40
- lerp_data = X_resampled.copy()
41
- lerp_data['multi_class'] = y_resampled
42
-
43
- # 제거변수 복구
44
- lerp_data['binary_class'] = lerp_data['multi_class'].apply(lambda x: 0 if x == 2 else 1)
45
- lerp_data['hour_sin'] = np.sin(2 * np.pi * lerp_data['hour'] / 24)
46
- lerp_data['hour_cos'] = np.cos(2 * np.pi * lerp_data['hour'] / 24)
47
- lerp_data['month_sin'] = np.sin(2 * np.pi * lerp_data['month'] / 12)
48
- lerp_data['month_cos'] = np.cos(2 * np.pi * lerp_data['month'] / 12)
49
- lerp_data['ground_temp - temp_C'] = lerp_data['groundtemp'] - lerp_data['temp_C']
50
-
51
- # 결과 저장
52
- lerp_data.to_csv(output_path, index = False)
53
- print(len(lerp_data[lerp_data['multi_class']==0]),'|',len(lerp_data[lerp_data['multi_class']==1]),'|',len(lerp_data[lerp_data['multi_class']==2]))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/make_train_test.ipynb DELETED
@@ -1,1099 +0,0 @@
1
- {
2
- "cells": [
3
- {
4
- "cell_type": "code",
5
- "execution_count": 1,
6
- "metadata": {},
7
- "outputs": [],
8
- "source": [
9
- "import pandas as pd\n",
10
- "import numpy as np\n",
11
- "import matplotlib.pyplot as plt\n",
12
- "import seaborn as sns\n",
13
- "from sklearn.model_selection import train_test_split\n",
14
- "from collections import Counter"
15
- ]
16
- },
17
- {
18
- "cell_type": "code",
19
- "execution_count": 2,
20
- "metadata": {},
21
- "outputs": [],
22
- "source": [
23
- "df_seoul = pd.read_feather(\"../data/data_for_modeling/df_seoul.feather\")\n",
24
- "df_busan = pd.read_feather(\"../data/data_for_modeling/df_busan.feather\")\n",
25
- "df_incheon = pd.read_feather(\"../data/data_for_modeling/df_incheon.feather\")\n",
26
- "df_daegu = pd.read_feather(\"../data/data_for_modeling/df_daegu.feather\")\n",
27
- "df_daejeon = pd.read_feather(\"../data/data_for_modeling/df_daejeon.feather\")\n",
28
- "df_gwangju = pd.read_feather(\"../data/data_for_modeling/df_gwangju.feather\")"
29
- ]
30
- },
31
- {
32
- "cell_type": "code",
33
- "execution_count": 3,
34
- "metadata": {},
35
- "outputs": [
36
- {
37
- "data": {
38
- "text/plain": [
39
- "Counter({2: 48534, 1: 3941, 0: 109})"
40
- ]
41
- },
42
- "execution_count": 3,
43
- "metadata": {},
44
- "output_type": "execute_result"
45
- }
46
- ],
47
- "source": [
48
- "Counter(df_seoul['multi_class'])"
49
- ]
50
- },
51
- {
52
- "cell_type": "code",
53
- "execution_count": 4,
54
- "metadata": {},
55
- "outputs": [
56
- {
57
- "data": {
58
- "text/plain": [
59
- "Counter({2: 50069, 1: 2350, 0: 165})"
60
- ]
61
- },
62
- "execution_count": 4,
63
- "metadata": {},
64
- "output_type": "execute_result"
65
- }
66
- ],
67
- "source": [
68
- "Counter(df_busan['multi_class'])"
69
- ]
70
- },
71
- {
72
- "cell_type": "code",
73
- "execution_count": 5,
74
- "metadata": {},
75
- "outputs": [
76
- {
77
- "data": {
78
- "text/plain": [
79
- "Counter({2: 44944, 1: 6658, 0: 982})"
80
- ]
81
- },
82
- "execution_count": 5,
83
- "metadata": {},
84
- "output_type": "execute_result"
85
- }
86
- ],
87
- "source": [
88
- "Counter(df_incheon['multi_class'])"
89
- ]
90
- },
91
- {
92
- "cell_type": "code",
93
- "execution_count": 6,
94
- "metadata": {},
95
- "outputs": [
96
- {
97
- "data": {
98
- "text/plain": [
99
- "Counter({2: 50919, 1: 1610, 0: 55})"
100
- ]
101
- },
102
- "execution_count": 6,
103
- "metadata": {},
104
- "output_type": "execute_result"
105
- }
106
- ],
107
- "source": [
108
- "Counter(df_daegu['multi_class'])"
109
- ]
110
- },
111
- {
112
- "cell_type": "code",
113
- "execution_count": 7,
114
- "metadata": {},
115
- "outputs": [
116
- {
117
- "data": {
118
- "text/plain": [
119
- "Counter({2: 48047, 1: 4227, 0: 310})"
120
- ]
121
- },
122
- "execution_count": 7,
123
- "metadata": {},
124
- "output_type": "execute_result"
125
- }
126
- ],
127
- "source": [
128
- "Counter(df_daejeon['multi_class'])"
129
- ]
130
- },
131
- {
132
- "cell_type": "code",
133
- "execution_count": 8,
134
- "metadata": {},
135
- "outputs": [
136
- {
137
- "data": {
138
- "text/plain": [
139
- "Counter({2: 48405, 1: 4015, 0: 164})"
140
- ]
141
- },
142
- "execution_count": 8,
143
- "metadata": {},
144
- "output_type": "execute_result"
145
- }
146
- ],
147
- "source": [
148
- "Counter(df_gwangju['multi_class'])"
149
- ]
150
- },
151
- {
152
- "cell_type": "code",
153
- "execution_count": 9,
154
- "metadata": {},
155
- "outputs": [
156
- {
157
- "data": {
158
- "text/plain": [
159
- "(52584, 30)"
160
- ]
161
- },
162
- "execution_count": 9,
163
- "metadata": {},
164
- "output_type": "execute_result"
165
- }
166
- ],
167
- "source": [
168
- "df_seoul.shape"
169
- ]
170
- },
171
- {
172
- "cell_type": "code",
173
- "execution_count": 10,
174
- "metadata": {},
175
- "outputs": [],
176
- "source": [
177
- "df_seoul = df_seoul.loc[df_seoul['year'].isin([2018, 2019, 2020, 2021]),:].copy()\n",
178
- "df_busan = df_busan.loc[df_busan['year'].isin([2018, 2019, 2020, 2021]),:].copy()\n",
179
- "df_incheon = df_incheon.loc[df_incheon['year'].isin([2018, 2019, 2020, 2021]),:].copy()\n",
180
- "df_daegu = df_daegu.loc[df_daegu['year'].isin([2018, 2019, 2020, 2021]),:].copy()\n",
181
- "df_daejeon = df_daejeon.loc[df_daejeon['year'].isin([2018, 2019, 2020, 2021]),:].copy()\n",
182
- "df_gwangju = df_gwangju.loc[df_gwangju['year'].isin([2018, 2019, 2020, 2021]),:].copy()"
183
- ]
184
- },
185
- {
186
- "cell_type": "code",
187
- "execution_count": 11,
188
- "metadata": {},
189
- "outputs": [],
190
- "source": [
191
- "cols = [col for col in df_seoul.columns if col != \"multi_class\"] + [\"multi_class\"]"
192
- ]
193
- },
194
- {
195
- "cell_type": "code",
196
- "execution_count": 12,
197
- "metadata": {},
198
- "outputs": [],
199
- "source": [
200
- "df_seoul = df_seoul[cols]\n",
201
- "df_busan = df_busan[cols]\n",
202
- "df_incheon = df_incheon[cols]\n",
203
- "df_daegu = df_daegu[cols]\n",
204
- "df_daejeon = df_daejeon[cols]\n",
205
- "df_gwangju = df_gwangju[cols]"
206
- ]
207
- },
208
- {
209
- "cell_type": "code",
210
- "execution_count": 13,
211
- "metadata": {},
212
- "outputs": [],
213
- "source": [
214
- "df_seoul_train = df_seoul.loc[df_seoul['year'].isin([2018, 2019, 2020]),:].copy()\n",
215
- "df_seoul_test = df_seoul.loc[df_seoul['year'].isin([2021]),:].copy()\n",
216
- "\n",
217
- "df_busan_train = df_busan.loc[df_busan['year'].isin([2018, 2019, 2020]),:].copy()\n",
218
- "df_busan_test = df_busan.loc[df_busan['year'].isin([2021]),:].copy()\n",
219
- "\n",
220
- "df_incheon_train = df_incheon.loc[df_incheon['year'].isin([2018, 2019, 2020]),:].copy()\n",
221
- "df_incheon_test = df_incheon.loc[df_incheon['year'].isin([2021]),:].copy()\n",
222
- "\n",
223
- "df_daegu_train = df_daegu.loc[df_daegu['year'].isin([2018, 2019, 2020]),:].copy()\n",
224
- "df_daegu_test = df_daegu.loc[df_daegu['year'].isin([2021]),:].copy()\n",
225
- "\n",
226
- "df_daejeon_train = df_daejeon.loc[df_daejeon['year'].isin([2018, 2019, 2020]),:].copy()\n",
227
- "df_daejeon_test = df_daejeon.loc[df_daejeon['year'].isin([2021]),:].copy()\n",
228
- "\n",
229
- "df_gwangju_train = df_gwangju.loc[df_gwangju['year'].isin([2018, 2019, 2020]),:].copy()\n",
230
- "df_gwangju_test = df_gwangju.loc[df_gwangju['year'].isin([2021]),:].copy()"
231
- ]
232
- },
233
- {
234
- "cell_type": "code",
235
- "execution_count": 14,
236
- "metadata": {},
237
- "outputs": [
238
- {
239
- "data": {
240
- "text/html": [
241
- "<div>\n",
242
- "<style scoped>\n",
243
- " .dataframe tbody tr th:only-of-type {\n",
244
- " vertical-align: middle;\n",
245
- " }\n",
246
- "\n",
247
- " .dataframe tbody tr th {\n",
248
- " vertical-align: top;\n",
249
- " }\n",
250
- "\n",
251
- " .dataframe thead th {\n",
252
- " text-align: right;\n",
253
- " }\n",
254
- "</style>\n",
255
- "<table border=\"1\" class=\"dataframe\">\n",
256
- " <thead>\n",
257
- " <tr style=\"text-align: right;\">\n",
258
- " <th></th>\n",
259
- " <th>temp_C</th>\n",
260
- " <th>precip_mm</th>\n",
261
- " <th>wind_speed</th>\n",
262
- " <th>wind_dir</th>\n",
263
- " <th>hm</th>\n",
264
- " <th>vap_pressure</th>\n",
265
- " <th>dewpoint_C</th>\n",
266
- " <th>loc_pressure</th>\n",
267
- " <th>sea_pressure</th>\n",
268
- " <th>solarRad</th>\n",
269
- " <th>...</th>\n",
270
- " <th>year</th>\n",
271
- " <th>month</th>\n",
272
- " <th>hour</th>\n",
273
- " <th>ground_temp - temp_C</th>\n",
274
- " <th>hour_sin</th>\n",
275
- " <th>hour_cos</th>\n",
276
- " <th>month_sin</th>\n",
277
- " <th>month_cos</th>\n",
278
- " <th>visi</th>\n",
279
- " <th>multi_class</th>\n",
280
- " </tr>\n",
281
- " </thead>\n",
282
- " <tbody>\n",
283
- " <tr>\n",
284
- " <th>0</th>\n",
285
- " <td>1.2</td>\n",
286
- " <td>0.0</td>\n",
287
- " <td>1.6</td>\n",
288
- " <td>360</td>\n",
289
- " <td>35.0</td>\n",
290
- " <td>2.3</td>\n",
291
- " <td>-12.6</td>\n",
292
- " <td>1015.8</td>\n",
293
- " <td>1024.6</td>\n",
294
- " <td>0.00</td>\n",
295
- " <td>...</td>\n",
296
- " <td>2018</td>\n",
297
- " <td>1</td>\n",
298
- " <td>0</td>\n",
299
- " <td>-5.4</td>\n",
300
- " <td>0.000000</td>\n",
301
- " <td>1.000000e+00</td>\n",
302
- " <td>0.5</td>\n",
303
- " <td>0.866025</td>\n",
304
- " <td>2000.0</td>\n",
305
- " <td>2</td>\n",
306
- " </tr>\n",
307
- " <tr>\n",
308
- " <th>1</th>\n",
309
- " <td>0.5</td>\n",
310
- " <td>0.0</td>\n",
311
- " <td>1.3</td>\n",
312
- " <td>360</td>\n",
313
- " <td>33.0</td>\n",
314
- " <td>2.1</td>\n",
315
- " <td>-13.9</td>\n",
316
- " <td>1015.5</td>\n",
317
- " <td>1024.3</td>\n",
318
- " <td>0.00</td>\n",
319
- " <td>...</td>\n",
320
- " <td>2018</td>\n",
321
- " <td>1</td>\n",
322
- " <td>1</td>\n",
323
- " <td>-5.4</td>\n",
324
- " <td>0.258819</td>\n",
325
- " <td>9.659258e-01</td>\n",
326
- " <td>0.5</td>\n",
327
- " <td>0.866025</td>\n",
328
- " <td>2000.0</td>\n",
329
- " <td>2</td>\n",
330
- " </tr>\n",
331
- " <tr>\n",
332
- " <th>2</th>\n",
333
- " <td>0.1</td>\n",
334
- " <td>0.0</td>\n",
335
- " <td>1.5</td>\n",
336
- " <td>20</td>\n",
337
- " <td>34.0</td>\n",
338
- " <td>2.1</td>\n",
339
- " <td>-13.9</td>\n",
340
- " <td>1015.7</td>\n",
341
- " <td>1024.5</td>\n",
342
- " <td>0.00</td>\n",
343
- " <td>...</td>\n",
344
- " <td>2018</td>\n",
345
- " <td>1</td>\n",
346
- " <td>2</td>\n",
347
- " <td>-5.4</td>\n",
348
- " <td>0.500000</td>\n",
349
- " <td>8.660254e-01</td>\n",
350
- " <td>0.5</td>\n",
351
- " <td>0.866025</td>\n",
352
- " <td>2000.0</td>\n",
353
- " <td>2</td>\n",
354
- " </tr>\n",
355
- " <tr>\n",
356
- " <th>3</th>\n",
357
- " <td>0.0</td>\n",
358
- " <td>0.0</td>\n",
359
- " <td>2.1</td>\n",
360
- " <td>320</td>\n",
361
- " <td>37.0</td>\n",
362
- " <td>2.3</td>\n",
363
- " <td>-12.9</td>\n",
364
- " <td>1015.9</td>\n",
365
- " <td>1024.7</td>\n",
366
- " <td>0.00</td>\n",
367
- " <td>...</td>\n",
368
- " <td>2018</td>\n",
369
- " <td>1</td>\n",
370
- " <td>3</td>\n",
371
- " <td>-5.0</td>\n",
372
- " <td>0.707107</td>\n",
373
- " <td>7.071068e-01</td>\n",
374
- " <td>0.5</td>\n",
375
- " <td>0.866025</td>\n",
376
- " <td>2000.0</td>\n",
377
- " <td>2</td>\n",
378
- " </tr>\n",
379
- " <tr>\n",
380
- " <th>4</th>\n",
381
- " <td>-0.1</td>\n",
382
- " <td>0.0</td>\n",
383
- " <td>2.3</td>\n",
384
- " <td>340</td>\n",
385
- " <td>42.0</td>\n",
386
- " <td>2.5</td>\n",
387
- " <td>-11.5</td>\n",
388
- " <td>1016.0</td>\n",
389
- " <td>1024.9</td>\n",
390
- " <td>0.00</td>\n",
391
- " <td>...</td>\n",
392
- " <td>2018</td>\n",
393
- " <td>1</td>\n",
394
- " <td>4</td>\n",
395
- " <td>-4.3</td>\n",
396
- " <td>0.866025</td>\n",
397
- " <td>5.000000e-01</td>\n",
398
- " <td>0.5</td>\n",
399
- " <td>0.866025</td>\n",
400
- " <td>2000.0</td>\n",
401
- " <td>2</td>\n",
402
- " </tr>\n",
403
- " <tr>\n",
404
- " <th>5</th>\n",
405
- " <td>-0.1</td>\n",
406
- " <td>0.0</td>\n",
407
- " <td>2.8</td>\n",
408
- " <td>50</td>\n",
409
- " <td>43.0</td>\n",
410
- " <td>2.6</td>\n",
411
- " <td>-11.2</td>\n",
412
- " <td>1016.0</td>\n",
413
- " <td>1024.9</td>\n",
414
- " <td>0.00</td>\n",
415
- " <td>...</td>\n",
416
- " <td>2018</td>\n",
417
- " <td>1</td>\n",
418
- " <td>5</td>\n",
419
- " <td>-4.0</td>\n",
420
- " <td>0.965926</td>\n",
421
- " <td>2.588190e-01</td>\n",
422
- " <td>0.5</td>\n",
423
- " <td>0.866025</td>\n",
424
- " <td>2000.0</td>\n",
425
- " <td>2</td>\n",
426
- " </tr>\n",
427
- " <tr>\n",
428
- " <th>6</th>\n",
429
- " <td>-0.5</td>\n",
430
- " <td>0.0</td>\n",
431
- " <td>2.1</td>\n",
432
- " <td>20</td>\n",
433
- " <td>45.0</td>\n",
434
- " <td>2.6</td>\n",
435
- " <td>-11.0</td>\n",
436
- " <td>1016.5</td>\n",
437
- " <td>1025.4</td>\n",
438
- " <td>0.00</td>\n",
439
- " <td>...</td>\n",
440
- " <td>2018</td>\n",
441
- " <td>1</td>\n",
442
- " <td>6</td>\n",
443
- " <td>-4.1</td>\n",
444
- " <td>1.000000</td>\n",
445
- " <td>6.123234e-17</td>\n",
446
- " <td>0.5</td>\n",
447
- " <td>0.866025</td>\n",
448
- " <td>2000.0</td>\n",
449
- " <td>2</td>\n",
450
- " </tr>\n",
451
- " <tr>\n",
452
- " <th>7</th>\n",
453
- " <td>-0.8</td>\n",
454
- " <td>0.0</td>\n",
455
- " <td>2.5</td>\n",
456
- " <td>340</td>\n",
457
- " <td>45.0</td>\n",
458
- " <td>2.6</td>\n",
459
- " <td>-11.2</td>\n",
460
- " <td>1017.1</td>\n",
461
- " <td>1026.0</td>\n",
462
- " <td>0.00</td>\n",
463
- " <td>...</td>\n",
464
- " <td>2018</td>\n",
465
- " <td>1</td>\n",
466
- " <td>7</td>\n",
467
- " <td>-4.5</td>\n",
468
- " <td>0.965926</td>\n",
469
- " <td>-2.588190e-01</td>\n",
470
- " <td>0.5</td>\n",
471
- " <td>0.866025</td>\n",
472
- " <td>2000.0</td>\n",
473
- " <td>2</td>\n",
474
- " </tr>\n",
475
- " <tr>\n",
476
- " <th>8</th>\n",
477
- " <td>-0.5</td>\n",
478
- " <td>0.0</td>\n",
479
- " <td>1.2</td>\n",
480
- " <td>360</td>\n",
481
- " <td>43.0</td>\n",
482
- " <td>2.5</td>\n",
483
- " <td>-11.5</td>\n",
484
- " <td>1017.4</td>\n",
485
- " <td>1026.3</td>\n",
486
- " <td>0.03</td>\n",
487
- " <td>...</td>\n",
488
- " <td>2018</td>\n",
489
- " <td>1</td>\n",
490
- " <td>8</td>\n",
491
- " <td>-4.0</td>\n",
492
- " <td>0.866025</td>\n",
493
- " <td>-5.000000e-01</td>\n",
494
- " <td>0.5</td>\n",
495
- " <td>0.866025</td>\n",
496
- " <td>2000.0</td>\n",
497
- " <td>2</td>\n",
498
- " </tr>\n",
499
- " <tr>\n",
500
- " <th>9</th>\n",
501
- " <td>1.7</td>\n",
502
- " <td>0.0</td>\n",
503
- " <td>2.1</td>\n",
504
- " <td>20</td>\n",
505
- " <td>39.0</td>\n",
506
- " <td>2.7</td>\n",
507
- " <td>-10.8</td>\n",
508
- " <td>1018.1</td>\n",
509
- " <td>1026.9</td>\n",
510
- " <td>0.46</td>\n",
511
- " <td>...</td>\n",
512
- " <td>2018</td>\n",
513
- " <td>1</td>\n",
514
- " <td>9</td>\n",
515
- " <td>2.8</td>\n",
516
- " <td>0.707107</td>\n",
517
- " <td>-7.071068e-01</td>\n",
518
- " <td>0.5</td>\n",
519
- " <td>0.866025</td>\n",
520
- " <td>1953.0</td>\n",
521
- " <td>2</td>\n",
522
- " </tr>\n",
523
- " </tbody>\n",
524
- "</table>\n",
525
- "<p>10 rows × 30 columns</p>\n",
526
- "</div>"
527
- ],
528
- "text/plain": [
529
- " temp_C precip_mm wind_speed wind_dir hm vap_pressure dewpoint_C \\\n",
530
- "0 1.2 0.0 1.6 360 35.0 2.3 -12.6 \n",
531
- "1 0.5 0.0 1.3 360 33.0 2.1 -13.9 \n",
532
- "2 0.1 0.0 1.5 20 34.0 2.1 -13.9 \n",
533
- "3 0.0 0.0 2.1 320 37.0 2.3 -12.9 \n",
534
- "4 -0.1 0.0 2.3 340 42.0 2.5 -11.5 \n",
535
- "5 -0.1 0.0 2.8 50 43.0 2.6 -11.2 \n",
536
- "6 -0.5 0.0 2.1 20 45.0 2.6 -11.0 \n",
537
- "7 -0.8 0.0 2.5 340 45.0 2.6 -11.2 \n",
538
- "8 -0.5 0.0 1.2 360 43.0 2.5 -11.5 \n",
539
- "9 1.7 0.0 2.1 20 39.0 2.7 -10.8 \n",
540
- "\n",
541
- " loc_pressure sea_pressure solarRad ... year month hour \\\n",
542
- "0 1015.8 1024.6 0.00 ... 2018 1 0 \n",
543
- "1 1015.5 1024.3 0.00 ... 2018 1 1 \n",
544
- "2 1015.7 1024.5 0.00 ... 2018 1 2 \n",
545
- "3 1015.9 1024.7 0.00 ... 2018 1 3 \n",
546
- "4 1016.0 1024.9 0.00 ... 2018 1 4 \n",
547
- "5 1016.0 1024.9 0.00 ... 2018 1 5 \n",
548
- "6 1016.5 1025.4 0.00 ... 2018 1 6 \n",
549
- "7 1017.1 1026.0 0.00 ... 2018 1 7 \n",
550
- "8 1017.4 1026.3 0.03 ... 2018 1 8 \n",
551
- "9 1018.1 1026.9 0.46 ... 2018 1 9 \n",
552
- "\n",
553
- " ground_temp - temp_C hour_sin hour_cos month_sin month_cos visi \\\n",
554
- "0 -5.4 0.000000 1.000000e+00 0.5 0.866025 2000.0 \n",
555
- "1 -5.4 0.258819 9.659258e-01 0.5 0.866025 2000.0 \n",
556
- "2 -5.4 0.500000 8.660254e-01 0.5 0.866025 2000.0 \n",
557
- "3 -5.0 0.707107 7.071068e-01 0.5 0.866025 2000.0 \n",
558
- "4 -4.3 0.866025 5.000000e-01 0.5 0.866025 2000.0 \n",
559
- "5 -4.0 0.965926 2.588190e-01 0.5 0.866025 2000.0 \n",
560
- "6 -4.1 1.000000 6.123234e-17 0.5 0.866025 2000.0 \n",
561
- "7 -4.5 0.965926 -2.588190e-01 0.5 0.866025 2000.0 \n",
562
- "8 -4.0 0.866025 -5.000000e-01 0.5 0.866025 2000.0 \n",
563
- "9 2.8 0.707107 -7.071068e-01 0.5 0.866025 1953.0 \n",
564
- "\n",
565
- " multi_class \n",
566
- "0 2 \n",
567
- "1 2 \n",
568
- "2 2 \n",
569
- "3 2 \n",
570
- "4 2 \n",
571
- "5 2 \n",
572
- "6 2 \n",
573
- "7 2 \n",
574
- "8 2 \n",
575
- "9 2 \n",
576
- "\n",
577
- "[10 rows x 30 columns]"
578
- ]
579
- },
580
- "execution_count": 14,
581
- "metadata": {},
582
- "output_type": "execute_result"
583
- }
584
- ],
585
- "source": [
586
- "df_busan_train.head(10)"
587
- ]
588
- },
589
- {
590
- "cell_type": "code",
591
- "execution_count": 15,
592
- "metadata": {},
593
- "outputs": [
594
- {
595
- "data": {
596
- "text/html": [
597
- "<div>\n",
598
- "<style scoped>\n",
599
- " .dataframe tbody tr th:only-of-type {\n",
600
- " vertical-align: middle;\n",
601
- " }\n",
602
- "\n",
603
- " .dataframe tbody tr th {\n",
604
- " vertical-align: top;\n",
605
- " }\n",
606
- "\n",
607
- " .dataframe thead th {\n",
608
- " text-align: right;\n",
609
- " }\n",
610
- "</style>\n",
611
- "<table border=\"1\" class=\"dataframe\">\n",
612
- " <thead>\n",
613
- " <tr style=\"text-align: right;\">\n",
614
- " <th></th>\n",
615
- " <th>temp_C</th>\n",
616
- " <th>precip_mm</th>\n",
617
- " <th>wind_speed</th>\n",
618
- " <th>wind_dir</th>\n",
619
- " <th>hm</th>\n",
620
- " <th>vap_pressure</th>\n",
621
- " <th>dewpoint_C</th>\n",
622
- " <th>loc_pressure</th>\n",
623
- " <th>sea_pressure</th>\n",
624
- " <th>solarRad</th>\n",
625
- " <th>...</th>\n",
626
- " <th>year</th>\n",
627
- " <th>month</th>\n",
628
- " <th>hour</th>\n",
629
- " <th>ground_temp - temp_C</th>\n",
630
- " <th>hour_sin</th>\n",
631
- " <th>hour_cos</th>\n",
632
- " <th>month_sin</th>\n",
633
- " <th>month_cos</th>\n",
634
- " <th>visi</th>\n",
635
- " <th>multi_class</th>\n",
636
- " </tr>\n",
637
- " </thead>\n",
638
- " <tbody>\n",
639
- " <tr>\n",
640
- " <th>26294</th>\n",
641
- " <td>0.1</td>\n",
642
- " <td>0.0</td>\n",
643
- " <td>6.3</td>\n",
644
- " <td>270</td>\n",
645
- " <td>37.0</td>\n",
646
- " <td>2.3</td>\n",
647
- " <td>-12.9</td>\n",
648
- " <td>1013.3</td>\n",
649
- " <td>1022.1</td>\n",
650
- " <td>2.07</td>\n",
651
- " <td>...</td>\n",
652
- " <td>2020</td>\n",
653
- " <td>12</td>\n",
654
- " <td>14</td>\n",
655
- " <td>5.8</td>\n",
656
- " <td>-0.500000</td>\n",
657
- " <td>-8.660254e-01</td>\n",
658
- " <td>-2.449294e-16</td>\n",
659
- " <td>1.0</td>\n",
660
- " <td>5000.0</td>\n",
661
- " <td>2</td>\n",
662
- " </tr>\n",
663
- " <tr>\n",
664
- " <th>26295</th>\n",
665
- " <td>1.2</td>\n",
666
- " <td>0.0</td>\n",
667
- " <td>5.9</td>\n",
668
- " <td>270</td>\n",
669
- " <td>35.0</td>\n",
670
- " <td>2.3</td>\n",
671
- " <td>-12.6</td>\n",
672
- " <td>1013.2</td>\n",
673
- " <td>1022.0</td>\n",
674
- " <td>1.71</td>\n",
675
- " <td>...</td>\n",
676
- " <td>2020</td>\n",
677
- " <td>12</td>\n",
678
- " <td>15</td>\n",
679
- " <td>5.6</td>\n",
680
- " <td>-0.707107</td>\n",
681
- " <td>-7.071068e-01</td>\n",
682
- " <td>-2.449294e-16</td>\n",
683
- " <td>1.0</td>\n",
684
- " <td>5000.0</td>\n",
685
- " <td>2</td>\n",
686
- " </tr>\n",
687
- " <tr>\n",
688
- " <th>26296</th>\n",
689
- " <td>1.6</td>\n",
690
- " <td>0.0</td>\n",
691
- " <td>3.6</td>\n",
692
- " <td>290</td>\n",
693
- " <td>34.0</td>\n",
694
- " <td>2.3</td>\n",
695
- " <td>-12.6</td>\n",
696
- " <td>1012.8</td>\n",
697
- " <td>1021.6</td>\n",
698
- " <td>1.14</td>\n",
699
- " <td>...</td>\n",
700
- " <td>2020</td>\n",
701
- " <td>12</td>\n",
702
- " <td>16</td>\n",
703
- " <td>1.4</td>\n",
704
- " <td>-0.866025</td>\n",
705
- " <td>-5.000000e-01</td>\n",
706
- " <td>-2.449294e-16</td>\n",
707
- " <td>1.0</td>\n",
708
- " <td>5000.0</td>\n",
709
- " <td>2</td>\n",
710
- " </tr>\n",
711
- " <tr>\n",
712
- " <th>26297</th>\n",
713
- " <td>1.2</td>\n",
714
- " <td>0.0</td>\n",
715
- " <td>3.8</td>\n",
716
- " <td>250</td>\n",
717
- " <td>38.0</td>\n",
718
- " <td>2.5</td>\n",
719
- " <td>-11.5</td>\n",
720
- " <td>1012.8</td>\n",
721
- " <td>1021.6</td>\n",
722
- " <td>0.48</td>\n",
723
- " <td>...</td>\n",
724
- " <td>2020</td>\n",
725
- " <td>12</td>\n",
726
- " <td>17</td>\n",
727
- " <td>-0.4</td>\n",
728
- " <td>-0.965926</td>\n",
729
- " <td>-2.588190e-01</td>\n",
730
- " <td>-2.449294e-16</td>\n",
731
- " <td>1.0</td>\n",
732
- " <td>5000.0</td>\n",
733
- " <td>2</td>\n",
734
- " </tr>\n",
735
- " <tr>\n",
736
- " <th>26298</th>\n",
737
- " <td>0.9</td>\n",
738
- " <td>0.0</td>\n",
739
- " <td>3.8</td>\n",
740
- " <td>270</td>\n",
741
- " <td>40.0</td>\n",
742
- " <td>2.6</td>\n",
743
- " <td>-11.2</td>\n",
744
- " <td>1013.1</td>\n",
745
- " <td>1021.9</td>\n",
746
- " <td>0.02</td>\n",
747
- " <td>...</td>\n",
748
- " <td>2020</td>\n",
749
- " <td>12</td>\n",
750
- " <td>18</td>\n",
751
- " <td>-0.8</td>\n",
752
- " <td>-1.000000</td>\n",
753
- " <td>-1.836970e-16</td>\n",
754
- " <td>-2.449294e-16</td>\n",
755
- " <td>1.0</td>\n",
756
- " <td>5000.0</td>\n",
757
- " <td>2</td>\n",
758
- " </tr>\n",
759
- " <tr>\n",
760
- " <th>26299</th>\n",
761
- " <td>0.6</td>\n",
762
- " <td>0.0</td>\n",
763
- " <td>6.2</td>\n",
764
- " <td>270</td>\n",
765
- " <td>41.0</td>\n",
766
- " <td>2.6</td>\n",
767
- " <td>-11.1</td>\n",
768
- " <td>1014.0</td>\n",
769
- " <td>1022.8</td>\n",
770
- " <td>0.00</td>\n",
771
- " <td>...</td>\n",
772
- " <td>2020</td>\n",
773
- " <td>12</td>\n",
774
- " <td>19</td>\n",
775
- " <td>-1.1</td>\n",
776
- " <td>-0.965926</td>\n",
777
- " <td>2.588190e-01</td>\n",
778
- " <td>-2.449294e-16</td>\n",
779
- " <td>1.0</td>\n",
780
- " <td>5000.0</td>\n",
781
- " <td>2</td>\n",
782
- " </tr>\n",
783
- " <tr>\n",
784
- " <th>26300</th>\n",
785
- " <td>0.1</td>\n",
786
- " <td>0.0</td>\n",
787
- " <td>6.0</td>\n",
788
- " <td>270</td>\n",
789
- " <td>44.0</td>\n",
790
- " <td>2.7</td>\n",
791
- " <td>-10.7</td>\n",
792
- " <td>1014.8</td>\n",
793
- " <td>1023.6</td>\n",
794
- " <td>0.00</td>\n",
795
- " <td>...</td>\n",
796
- " <td>2020</td>\n",
797
- " <td>12</td>\n",
798
- " <td>20</td>\n",
799
- " <td>-0.9</td>\n",
800
- " <td>-0.866025</td>\n",
801
- " <td>5.000000e-01</td>\n",
802
- " <td>-2.449294e-16</td>\n",
803
- " <td>1.0</td>\n",
804
- " <td>5000.0</td>\n",
805
- " <td>2</td>\n",
806
- " </tr>\n",
807
- " <tr>\n",
808
- " <th>26301</th>\n",
809
- " <td>-0.2</td>\n",
810
- " <td>0.0</td>\n",
811
- " <td>5.0</td>\n",
812
- " <td>290</td>\n",
813
- " <td>48.0</td>\n",
814
- " <td>2.9</td>\n",
815
- " <td>-9.9</td>\n",
816
- " <td>1014.6</td>\n",
817
- " <td>1023.4</td>\n",
818
- " <td>0.00</td>\n",
819
- " <td>...</td>\n",
820
- " <td>2020</td>\n",
821
- " <td>12</td>\n",
822
- " <td>21</td>\n",
823
- " <td>-0.8</td>\n",
824
- " <td>-0.707107</td>\n",
825
- " <td>7.071068e-01</td>\n",
826
- " <td>-2.449294e-16</td>\n",
827
- " <td>1.0</td>\n",
828
- " <td>5000.0</td>\n",
829
- " <td>2</td>\n",
830
- " </tr>\n",
831
- " <tr>\n",
832
- " <th>26302</th>\n",
833
- " <td>-0.7</td>\n",
834
- " <td>0.0</td>\n",
835
- " <td>2.7</td>\n",
836
- " <td>270</td>\n",
837
- " <td>51.0</td>\n",
838
- " <td>3.0</td>\n",
839
- " <td>-9.6</td>\n",
840
- " <td>1014.8</td>\n",
841
- " <td>1023.6</td>\n",
842
- " <td>0.00</td>\n",
843
- " <td>...</td>\n",
844
- " <td>2020</td>\n",
845
- " <td>12</td>\n",
846
- " <td>22</td>\n",
847
- " <td>-0.6</td>\n",
848
- " <td>-0.500000</td>\n",
849
- " <td>8.660254e-01</td>\n",
850
- " <td>-2.449294e-16</td>\n",
851
- " <td>1.0</td>\n",
852
- " <td>5000.0</td>\n",
853
- " <td>2</td>\n",
854
- " </tr>\n",
855
- " <tr>\n",
856
- " <th>26303</th>\n",
857
- " <td>-0.7</td>\n",
858
- " <td>0.0</td>\n",
859
- " <td>3.8</td>\n",
860
- " <td>250</td>\n",
861
- " <td>55.0</td>\n",
862
- " <td>3.2</td>\n",
863
- " <td>-8.6</td>\n",
864
- " <td>1015.1</td>\n",
865
- " <td>1024.0</td>\n",
866
- " <td>0.00</td>\n",
867
- " <td>...</td>\n",
868
- " <td>2020</td>\n",
869
- " <td>12</td>\n",
870
- " <td>23</td>\n",
871
- " <td>-0.6</td>\n",
872
- " <td>-0.258819</td>\n",
873
- " <td>9.659258e-01</td>\n",
874
- " <td>-2.449294e-16</td>\n",
875
- " <td>1.0</td>\n",
876
- " <td>5000.0</td>\n",
877
- " <td>2</td>\n",
878
- " </tr>\n",
879
- " </tbody>\n",
880
- "</table>\n",
881
- "<p>10 rows × 30 columns</p>\n",
882
- "</div>"
883
- ],
884
- "text/plain": [
885
- " temp_C precip_mm wind_speed wind_dir hm vap_pressure dewpoint_C \\\n",
886
- "26294 0.1 0.0 6.3 270 37.0 2.3 -12.9 \n",
887
- "26295 1.2 0.0 5.9 270 35.0 2.3 -12.6 \n",
888
- "26296 1.6 0.0 3.6 290 34.0 2.3 -12.6 \n",
889
- "26297 1.2 0.0 3.8 250 38.0 2.5 -11.5 \n",
890
- "26298 0.9 0.0 3.8 270 40.0 2.6 -11.2 \n",
891
- "26299 0.6 0.0 6.2 270 41.0 2.6 -11.1 \n",
892
- "26300 0.1 0.0 6.0 270 44.0 2.7 -10.7 \n",
893
- "26301 -0.2 0.0 5.0 290 48.0 2.9 -9.9 \n",
894
- "26302 -0.7 0.0 2.7 270 51.0 3.0 -9.6 \n",
895
- "26303 -0.7 0.0 3.8 250 55.0 3.2 -8.6 \n",
896
- "\n",
897
- " loc_pressure sea_pressure solarRad ... year month hour \\\n",
898
- "26294 1013.3 1022.1 2.07 ... 2020 12 14 \n",
899
- "26295 1013.2 1022.0 1.71 ... 2020 12 15 \n",
900
- "26296 1012.8 1021.6 1.14 ... 2020 12 16 \n",
901
- "26297 1012.8 1021.6 0.48 ... 2020 12 17 \n",
902
- "26298 1013.1 1021.9 0.02 ... 2020 12 18 \n",
903
- "26299 1014.0 1022.8 0.00 ... 2020 12 19 \n",
904
- "26300 1014.8 1023.6 0.00 ... 2020 12 20 \n",
905
- "26301 1014.6 1023.4 0.00 ... 2020 12 21 \n",
906
- "26302 1014.8 1023.6 0.00 ... 2020 12 22 \n",
907
- "26303 1015.1 1024.0 0.00 ... 2020 12 23 \n",
908
- "\n",
909
- " ground_temp - temp_C hour_sin hour_cos month_sin month_cos \\\n",
910
- "26294 5.8 -0.500000 -8.660254e-01 -2.449294e-16 1.0 \n",
911
- "26295 5.6 -0.707107 -7.071068e-01 -2.449294e-16 1.0 \n",
912
- "26296 1.4 -0.866025 -5.000000e-01 -2.449294e-16 1.0 \n",
913
- "26297 -0.4 -0.965926 -2.588190e-01 -2.449294e-16 1.0 \n",
914
- "26298 -0.8 -1.000000 -1.836970e-16 -2.449294e-16 1.0 \n",
915
- "26299 -1.1 -0.965926 2.588190e-01 -2.449294e-16 1.0 \n",
916
- "26300 -0.9 -0.866025 5.000000e-01 -2.449294e-16 1.0 \n",
917
- "26301 -0.8 -0.707107 7.071068e-01 -2.449294e-16 1.0 \n",
918
- "26302 -0.6 -0.500000 8.660254e-01 -2.449294e-16 1.0 \n",
919
- "26303 -0.6 -0.258819 9.659258e-01 -2.449294e-16 1.0 \n",
920
- "\n",
921
- " visi multi_class \n",
922
- "26294 5000.0 2 \n",
923
- "26295 5000.0 2 \n",
924
- "26296 5000.0 2 \n",
925
- "26297 5000.0 2 \n",
926
- "26298 5000.0 2 \n",
927
- "26299 5000.0 2 \n",
928
- "26300 5000.0 2 \n",
929
- "26301 5000.0 2 \n",
930
- "26302 5000.0 2 \n",
931
- "26303 5000.0 2 \n",
932
- "\n",
933
- "[10 rows x 30 columns]"
934
- ]
935
- },
936
- "execution_count": 15,
937
- "metadata": {},
938
- "output_type": "execute_result"
939
- }
940
- ],
941
- "source": [
942
- "df_busan_train.tail(10)"
943
- ]
944
- },
945
- {
946
- "cell_type": "code",
947
- "execution_count": 16,
948
- "metadata": {},
949
- "outputs": [
950
- {
951
- "name": "stdout",
952
- "output_type": "stream",
953
- "text": [
954
- "<class 'pandas.core.frame.DataFrame'>\n",
955
- "Index: 26304 entries, 0 to 26303\n",
956
- "Data columns (total 30 columns):\n",
957
- " # Column Non-Null Count Dtype \n",
958
- "--- ------ -------------- ----- \n",
959
- " 0 temp_C 26304 non-null float64 \n",
960
- " 1 precip_mm 26304 non-null float64 \n",
961
- " 2 wind_speed 26304 non-null float64 \n",
962
- " 3 wind_dir 26304 non-null category\n",
963
- " 4 hm 26304 non-null float64 \n",
964
- " 5 vap_pressure 26304 non-null float64 \n",
965
- " 6 dewpoint_C 26304 non-null float64 \n",
966
- " 7 loc_pressure 26304 non-null float64 \n",
967
- " 8 sea_pressure 26304 non-null float64 \n",
968
- " 9 solarRad 26304 non-null float64 \n",
969
- " 10 snow_cm 26304 non-null float64 \n",
970
- " 11 cloudcover 26304 non-null category\n",
971
- " 12 lm_cloudcover 26304 non-null category\n",
972
- " 13 low_cloudbase 26304 non-null float64 \n",
973
- " 14 groundtemp 26304 non-null float64 \n",
974
- " 15 O3 26304 non-null float64 \n",
975
- " 16 NO2 26304 non-null float64 \n",
976
- " 17 PM10 26304 non-null float64 \n",
977
- " 18 PM25 26304 non-null float64 \n",
978
- " 19 binary_class 26304 non-null int64 \n",
979
- " 20 year 26304 non-null int64 \n",
980
- " 21 month 26304 non-null int64 \n",
981
- " 22 hour 26304 non-null int64 \n",
982
- " 23 ground_temp - temp_C 26304 non-null float64 \n",
983
- " 24 hour_sin 26304 non-null float64 \n",
984
- " 25 hour_cos 26304 non-null float64 \n",
985
- " 26 month_sin 26304 non-null float64 \n",
986
- " 27 month_cos 26304 non-null float64 \n",
987
- " 28 visi 26304 non-null float64 \n",
988
- " 29 multi_class 26304 non-null int64 \n",
989
- "dtypes: category(3), float64(22), int64(5)\n",
990
- "memory usage: 5.7 MB\n"
991
- ]
992
- }
993
- ],
994
- "source": [
995
- "df_busan_train.info()"
996
- ]
997
- },
998
- {
999
- "cell_type": "code",
1000
- "execution_count": 17,
1001
- "metadata": {},
1002
- "outputs": [],
1003
- "source": [
1004
- "df_seoul_train.to_csv(\"../data/data_for_modeling/seoul_train.csv\")\n",
1005
- "df_seoul_test.to_csv(\"../data/data_for_modeling/seoul_test.csv\")\n",
1006
- "\n",
1007
- "df_busan_train.to_csv(\"../data/data_for_modeling/busan_train.csv\")\n",
1008
- "df_busan_test.to_csv(\"../data/data_for_modeling/busan_test.csv\")\n",
1009
- "\n",
1010
- "df_incheon_train.to_csv(\"../data/data_for_modeling/incheon_train.csv\")\n",
1011
- "df_incheon_test.to_csv(\"../data/data_for_modeling/incheon_test.csv\")\n",
1012
- "\n",
1013
- "df_daegu_train.to_csv(\"../data/data_for_modeling/daegu_train.csv\")\n",
1014
- "df_daegu_test.to_csv(\"../data/data_for_modeling/daegu_test.csv\")\n",
1015
- "\n",
1016
- "df_daejeon_train.to_csv(\"../data/data_for_modeling/daejeon_train.csv\")\n",
1017
- "df_daejeon_test.to_csv(\"../data/data_for_modeling/daejeon_test.csv\")\n",
1018
- "\n",
1019
- "df_gwangju_train.to_csv(\"../data/data_for_modeling/gwangju_train.csv\")\n",
1020
- "df_gwangju_test.to_csv(\"../data/data_for_modeling/gwangju_test.csv\")\n",
1021
- "\n",
1022
- "df_seoul_train = pd.read_csv(\"../data/data_for_modeling/seoul_train.csv\")\n",
1023
- "df_seoul_test = pd.read_csv(\"../data/data_for_modeling/seoul_test.csv\")\n"
1024
- ]
1025
- },
1026
- {
1027
- "cell_type": "code",
1028
- "execution_count": 18,
1029
- "metadata": {},
1030
- "outputs": [
1031
- {
1032
- "name": "stdout",
1033
- "output_type": "stream",
1034
- "text": [
1035
- "Counter({2: 8266, 1: 481, 0: 13})\n",
1036
- "Counter({2: 23686, 1: 2579, 0: 39})\n",
1037
- "Counter({2: 8455, 1: 281, 0: 24})\n",
1038
- "Counter({2: 24694, 1: 1516, 0: 94})\n",
1039
- "Counter({2: 7373, 1: 1205, 0: 182})\n",
1040
- "Counter({2: 21893, 1: 3892, 0: 519})\n",
1041
- "Counter({2: 8631, 1: 128, 0: 1})\n",
1042
- "Counter({2: 25149, 1: 1107, 0: 48})\n",
1043
- "Counter({2: 8089, 1: 618, 0: 53})\n",
1044
- "Counter({2: 23471, 1: 2660, 0: 173})\n",
1045
- "Counter({2: 8087, 1: 643, 0: 30})\n",
1046
- "Counter({2: 23798, 1: 2411, 0: 95})\n"
1047
- ]
1048
- }
1049
- ],
1050
- "source": [
1051
- "print(Counter(df_seoul_test['multi_class']))\n",
1052
- "print(Counter(df_seoul_train['multi_class']))\n",
1053
- "\n",
1054
- "print(Counter(df_busan_test['multi_class']))\n",
1055
- "print(Counter(df_busan_train['multi_class']))\n",
1056
- "\n",
1057
- "print(Counter(df_incheon_test['multi_class']))\n",
1058
- "print(Counter(df_incheon_train['multi_class']))\n",
1059
- "\n",
1060
- "print(Counter(df_daegu_test['multi_class']))\n",
1061
- "print(Counter(df_daegu_train['multi_class']))\n",
1062
- "\n",
1063
- "print(Counter(df_daejeon_test['multi_class']))\n",
1064
- "print(Counter(df_daejeon_train['multi_class']))\n",
1065
- "\n",
1066
- "print(Counter(df_gwangju_test['multi_class']))\n",
1067
- "print(Counter(df_gwangju_train['multi_class']))"
1068
- ]
1069
- },
1070
- {
1071
- "cell_type": "code",
1072
- "execution_count": null,
1073
- "metadata": {},
1074
- "outputs": [],
1075
- "source": []
1076
- }
1077
- ],
1078
- "metadata": {
1079
- "kernelspec": {
1080
- "display_name": "Python 3",
1081
- "language": "python",
1082
- "name": "python3"
1083
- },
1084
- "language_info": {
1085
- "codemirror_mode": {
1086
- "name": "ipython",
1087
- "version": 3
1088
- },
1089
- "file_extension": ".py",
1090
- "mimetype": "text/x-python",
1091
- "name": "python",
1092
- "nbconvert_exporter": "python",
1093
- "pygments_lexer": "ipython3",
1094
- "version": "3.8.10"
1095
- }
1096
- },
1097
- "nbformat": 4,
1098
- "nbformat_minor": 2
1099
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/model_result/best_sample/ensemble_best_sample.csv DELETED
@@ -1,157 +0,0 @@
1
- model,CSI,MCC,Accuracy,region,data_sample
2
- deepgbm+ft_transformer,0.6992424754332419,0.8049778134563997,0.9617050719032529,seoul,best
3
- deepgbm+resnet_like,0.7090721222546449,0.8134996041947694,0.9639575067994112,seoul,best
4
- deepgbm+XGBoost,0.6692268948642789,0.7828377071996622,0.9579950262411524,seoul,best
5
- deepgbm+LightGBM,0.6935149238495999,0.801572479004747,0.9620935948299524,seoul,best
6
- ft_transformer+resnet_like,0.6252731559795573,0.7509360530096033,0.9488043848924154,seoul,best
7
- ft_transformer+XGBoost,0.6142839547044981,0.7403116223696319,0.9470938235563208,seoul,best
8
- ft_transformer+LightGBM,0.6085686211212852,0.7359331482423171,0.9466367866856302,seoul,best
9
- resnet_like+XGBoost,0.6194270544060083,0.7446567256109252,0.9478222130731675,seoul,best
10
- resnet_like+LightGBM,0.6136847718241535,0.741095701007121,0.9457727208457053,seoul,best
11
- XGBoost+LightGBM,0.5861106806341969,0.7149062718037028,0.9421973118413721,seoul,best
12
- deepgbm+ft_transformer+resnet_like,0.6821685259093041,0.7946632359980877,0.9596256373148356,seoul,best
13
- deepgbm+ft_transformer+XGBoost,0.6730333196442752,0.7867573605932577,0.9584481123836616,seoul,best
14
- deepgbm+ft_transformer+LightGBM,0.6764764169794865,0.7902017631994056,0.9590548610591277,seoul,best
15
- deepgbm+resnet_like+XGBoost,0.6784375221650579,0.7915054245115618,0.9591382422170655,seoul,best
16
- deepgbm+resnet_like+LightGBM,0.6775625748710835,0.7913206958685782,0.9590235671332685,seoul,best
17
- deepgbm+XGBoost+LightGBM,0.6325378315490732,0.7547274504003308,0.9511999817018905,seoul,best
18
- ft_transformer+resnet_like+XGBoost,0.6332314687362993,0.7561343704178952,0.9500612362868145,seoul,best
19
- ft_transformer+resnet_like+LightGBM,0.6306548804840734,0.7545886604473321,0.9496810306826026,seoul,best
20
- ft_transformer+XGBoost+LightGBM,0.6052879958628536,0.7319675532436248,0.9455772637672482,seoul,best
21
- resnet_like+XGBoost+LightGBM,0.6103490502864374,0.7369455626993492,0.9460374196338716,seoul,best
22
- deepgbm+ft_transformer+resnet_like+XGBoost,0.6729967545904317,0.7875845379790564,0.9581108449567916,seoul,best
23
- deepgbm+ft_transformer+resnet_like+LightGBM,0.6719561859986568,0.7873723840805852,0.9580337017907196,seoul,best
24
- deepgbm+ft_transformer+XGBoost+LightGBM,0.6531335492225346,0.7716405213564462,0.9549966314843926,seoul,best
25
- deepgbm+resnet_like+XGBoost+LightGBM,0.6495283344540792,0.769203226760809,0.9539740166845488,seoul,best
26
- ft_transformer+resnet_like+XGBoost+LightGBM,0.6233313312481723,0.7475113353931637,0.9482768587136429,seoul,best
27
- deepgbm+ft_transformer+resnet_like+XGBoost+LightGBM,0.6569118107203898,0.7752226196801869,0.955377564854322,seoul,best
28
- deepgbm+ft_transformer,0.6231286091598959,0.762319461740938,0.9692817368232819,busan,best
29
- deepgbm+resnet_like,0.693965736363066,0.8126142282014271,0.9774987316083207,busan,best
30
- deepgbm+XGBoost,0.5949465703841524,0.7372757652859034,0.9695935324500337,busan,best
31
- deepgbm+LightGBM,0.6134672160989406,0.7540785415035627,0.969397659505452,busan,best
32
- ft_transformer+resnet_like,0.5993289393140034,0.7415869283802814,0.9687921584283586,busan,best
33
- ft_transformer+XGBoost,0.5275703861594696,0.6836131837976293,0.9627103242924039,busan,best
34
- ft_transformer+LightGBM,0.5340117437566735,0.690570910687501,0.9611092397135513,busan,best
35
- resnet_like+XGBoost,0.5476348496496576,0.6975679320922632,0.9653371426670326,busan,best
36
- resnet_like+LightGBM,0.5541563454462936,0.7052881357398278,0.9637749415708096,busan,best
37
- XGBoost+LightGBM,0.4789290487253062,0.6395338303931094,0.9572717310843294,busan,best
38
- deepgbm+ft_transformer+resnet_like,0.6446887106206897,0.777619884223958,0.9728593291247681,busan,best
39
- deepgbm+ft_transformer+XGBoost,0.6102029002552042,0.7511249566662123,0.9699326712744633,busan,best
40
- deepgbm+ft_transformer+LightGBM,0.6162985577163776,0.7570233197218496,0.9692066729878318,busan,best
41
- deepgbm+resnet_like+XGBoost,0.6261344637257955,0.7612451856135231,0.9726364248821019,busan,best
42
- deepgbm+resnet_like+LightGBM,0.6357139958459742,0.770523881133966,0.9720613859986192,busan,best
43
- deepgbm+XGBoost+LightGBM,0.550954042564896,0.7024853673988001,0.9645722608977718,busan,best
44
- ft_transformer+resnet_like+XGBoost,0.5708235070191732,0.7183051671189992,0.9673502466086118,busan,best
45
- ft_transformer+resnet_like+LightGBM,0.5747801644059322,0.723481589964096,0.9662444585838926,busan,best
46
- ft_transformer+XGBoost+LightGBM,0.5188547215134367,0.6758426217880089,0.9612265139606259,busan,best
47
- resnet_like+XGBoost+LightGBM,0.5283032797443571,0.6828500944597963,0.9625962730077933,busan,best
48
- deepgbm+ft_transformer+resnet_like+XGBoost,0.62252465197324,0.7595480882291313,0.9716073641573305,busan,best
49
- deepgbm+ft_transformer+resnet_like+LightGBM,0.6273921955959187,0.7650978772099609,0.9711861957398673,busan,best
50
- deepgbm+ft_transformer+XGBoost+LightGBM,0.5803171388664561,0.7273427892946641,0.9671579085260872,busan,best
51
- deepgbm+resnet_like+XGBoost+LightGBM,0.5883015315947713,0.7322310098800981,0.9685279794728481,busan,best
52
- ft_transformer+resnet_like+XGBoost+LightGBM,0.5527408442450511,0.7040063830266932,0.9648394548826841,busan,best
53
- deepgbm+ft_transformer+resnet_like+XGBoost+LightGBM,0.5949050119452819,0.7385090163673939,0.9689076652444045,busan,best
54
- deepgbm+ft_transformer,0.5873884001633557,0.7086005084355499,0.9163964576523526,incheon,best
55
- deepgbm+resnet_like,0.5938343436639008,0.7129218344958757,0.9139223661119011,incheon,best
56
- deepgbm+XGBoost,0.5919031180535835,0.7111840783628853,0.9141871688665986,incheon,best
57
- deepgbm+LightGBM,0.5936054700869063,0.71280494431202,0.9151763064434298,incheon,best
58
- ft_transformer+resnet_like,0.5967079690105518,0.7167701525416347,0.9161293676339713,incheon,best
59
- ft_transformer+XGBoost,0.5958609493419124,0.7170391565071776,0.9169653625105005,incheon,best
60
- ft_transformer+LightGBM,0.5970916463252486,0.7186247751024871,0.9177277490830152,incheon,best
61
- resnet_like+XGBoost,0.6048691059122082,0.7230658352305586,0.9150990593108267,incheon,best
62
- resnet_like+LightGBM,0.6006642978391444,0.7204725929547785,0.913995974415916,incheon,best
63
- XGBoost+LightGBM,0.5923037998052801,0.7130252479770401,0.9127417221847942,incheon,best
64
- deepgbm+ft_transformer+resnet_like,0.5991788818331828,0.7177110572369007,0.9171559331619964,incheon,best
65
- deepgbm+ft_transformer+XGBoost,0.5957796816002722,0.7161817507572641,0.9171555172958721,incheon,best
66
- deepgbm+ft_transformer+LightGBM,0.5956266359998414,0.7160886468480842,0.9173455681147126,incheon,best
67
- deepgbm+resnet_like+XGBoost,0.606730149793507,0.7239728449518917,0.9167729204614451,incheon,best
68
- deepgbm+resnet_like+LightGBM,0.6063493089884,0.7235778913430906,0.9167717768296031,incheon,best
69
- deepgbm+XGBoost+LightGBM,0.6018765631426876,0.7202419676609231,0.9160501451372774,incheon,best
70
- ft_transformer+resnet_like+XGBoost,0.6044191487414033,0.7234603140558615,0.9175339554690555,incheon,best
71
- ft_transformer+resnet_like+LightGBM,0.6061367487233379,0.7248883683993704,0.9179906804401528,incheon,best
72
- ft_transformer+XGBoost+LightGBM,0.6013258447351273,0.721556226775587,0.9174587876670742,incheon,best
73
- resnet_like+XGBoost+LightGBM,0.6028836532105156,0.7216380012607432,0.915023475642721,incheon,best
74
- deepgbm+ft_transformer+resnet_like+XGBoost,0.6058688939038325,0.7235044499077539,0.9180666799743826,incheon,best
75
- deepgbm+ft_transformer+resnet_like+LightGBM,0.6053266964324343,0.7232391120300768,0.9180279004582844,incheon,best
76
- deepgbm+ft_transformer+XGBoost+LightGBM,0.6030917448851759,0.7220222485555309,0.9181051475908876,incheon,best
77
- deepgbm+resnet_like+XGBoost+LightGBM,0.6054312862809185,0.7231150429381622,0.9164303507414893,incheon,best
78
- ft_transformer+resnet_like+XGBoost+LightGBM,0.6046920079336132,0.7233826726564859,0.9171152822483387,incheon,best
79
- deepgbm+ft_transformer+resnet_like+XGBoost+LightGBM,0.6109420059115421,0.72786127635734,0.9189396869359815,incheon,best
80
- deepgbm+ft_transformer,0.6446005403603131,0.7684243711794304,0.9801833553742378,daegu,best
81
- deepgbm+resnet_like,0.6137498386585009,0.7496378336201855,0.9772962048057489,daegu,best
82
- deepgbm+XGBoost,0.5974990513708366,0.7398559419115336,0.9765007568763463,daegu,best
83
- deepgbm+LightGBM,0.6115123902876293,0.7497436182930696,0.9776771381756785,daegu,best
84
- ft_transformer+resnet_like,0.5469112144188149,0.6944690161411611,0.9743305595062838,daegu,best
85
- ft_transformer+XGBoost,0.5161941376039343,0.6722802482596641,0.9719755096439354,daegu,best
86
- ft_transformer+LightGBM,0.5073005220098658,0.6635248688295593,0.9720125217290049,daegu,best
87
- resnet_like+XGBoost,0.5029516970497304,0.6645749569024574,0.9683687027472115,daegu,best
88
- resnet_like+LightGBM,0.4951131290074195,0.6589965524803283,0.968139976378804,daegu,best
89
- XGBoost+LightGBM,0.4549973810817563,0.6238297325596118,0.9647562816578087,daegu,best
90
- deepgbm+ft_transformer+resnet_like,0.6183060943020874,0.7512947555661068,0.978549833237684,daegu,best
91
- deepgbm+ft_transformer+XGBoost,0.6017176745611856,0.7392748101693137,0.9775999950096065,daegu,best
92
- deepgbm+ft_transformer+LightGBM,0.6004996462619208,0.7388281525299276,0.9778661493292079,daegu,best
93
- deepgbm+resnet_like+XGBoost,0.5739664154870325,0.7226992953385828,0.9747144039390839,daegu,best
94
- deepgbm+resnet_like+LightGBM,0.5760590538958189,0.7240729842389874,0.975132349394083,daegu,best
95
- deepgbm+XGBoost+LightGBM,0.5273980786589871,0.6866104535247874,0.9714085801498781,daegu,best
96
- ft_transformer+resnet_like+XGBoost,0.5287822236272242,0.6819314770838893,0.9723570668130517,daegu,best
97
- ft_transformer+resnet_like+LightGBM,0.520853018041326,0.675931210971238,0.972014705026158,daegu,best
98
- ft_transformer+XGBoost+LightGBM,0.48437190164347993,0.6467498445011408,0.9690512430238457,daegu,best
99
- resnet_like+XGBoost+LightGBM,0.4803327637605557,0.6466009408894583,0.9668476723973018,daegu,best
100
- deepgbm+ft_transformer+resnet_like+XGBoost,0.5787636584539054,0.7225361676622583,0.9760432001730003,daegu,best
101
- deepgbm+ft_transformer+resnet_like+LightGBM,0.5827514275296815,0.7252407488954487,0.9763842103949898,daegu,best
102
- deepgbm+ft_transformer+XGBoost+LightGBM,0.5563231124331217,0.7066219754268251,0.9745609493391888,daegu,best
103
- deepgbm+resnet_like+XGBoost+LightGBM,0.5497033929377105,0.7036672133164642,0.9729662067187331,daegu,best
104
- ft_transformer+resnet_like+XGBoost+LightGBM,0.5057939271151667,0.6644447331953721,0.9703813908226664,daegu,best
105
- deepgbm+ft_transformer+resnet_like+XGBoost+LightGBM,0.5538086053234044,0.7036101925279703,0.9742201470502616,daegu,best
106
- deepgbm+ft_transformer,0.6798477338109503,0.7890627213525137,0.9581849730934616,daejeon,best
107
- deepgbm+resnet_like,0.665680366173557,0.7799054500657331,0.9548432808510284,daejeon,best
108
- deepgbm+XGBoost,0.6574065703012586,0.7732790182043394,0.9541580374445858,daejeon,best
109
- deepgbm+LightGBM,0.6615432493953055,0.7763462848232718,0.9551088113714433,daejeon,best
110
- ft_transformer+resnet_like,0.5874777029243061,0.7124933721407519,0.9448812078415716,daejeon,best
111
- ft_transformer+XGBoost,0.5819133878684225,0.7079795420327318,0.9433591378263508,daejeon,best
112
- ft_transformer+LightGBM,0.5805334443959108,0.7070987268766625,0.9437394473970939,daejeon,best
113
- resnet_like+XGBoost,0.5621088257073251,0.6925993749297947,0.9389891126248638,daejeon,best
114
- resnet_like+LightGBM,0.5552708987061287,0.6862528661139199,0.9386090109871829,daejeon,best
115
- XGBoost+LightGBM,0.5288440675809284,0.6634738950206458,0.9327136928080112,daejeon,best
116
- deepgbm+ft_transformer+resnet_like,0.6605998838783642,0.7743701608802702,0.9553335870116691,daejeon,best
117
- deepgbm+ft_transformer+XGBoost,0.6495239084733958,0.7651993867637442,0.9536631567565769,daejeon,best
118
- deepgbm+ft_transformer+LightGBM,0.6556854096497969,0.7699591127202252,0.9550316682053713,daejeon,best
119
- deepgbm+resnet_like+XGBoost,0.646377396487214,0.7644362437422076,0.9526370070946761,daejeon,best
120
- deepgbm+resnet_like+LightGBM,0.6441717942778256,0.7623630196443353,0.9527517861450042,daejeon,best
121
- deepgbm+XGBoost+LightGBM,0.6104936580055047,0.7355161749128437,0.9470099225657277,daejeon,best
122
- ft_transformer+resnet_like+XGBoost,0.5926724219238438,0.7169506517948077,0.9453380367792,daejeon,best
123
- ft_transformer+resnet_like+LightGBM,0.5868504053718506,0.7120674745305696,0.9448050003742795,daejeon,best
124
- ft_transformer+XGBoost+LightGBM,0.5697749480416849,0.6978540261423789,0.9408133093794446,daejeon,best
125
- resnet_like+XGBoost+LightGBM,0.5568421888456356,0.6879639430352676,0.9383421289018639,daejeon,best
126
- deepgbm+ft_transformer+resnet_like+XGBoost,0.6464484806055241,0.7628617034260374,0.953282847185834,daejeon,best
127
- deepgbm+ft_transformer+resnet_like+LightGBM,0.6432761028976647,0.7603027833589299,0.9530166928662326,daejeon,best
128
- deepgbm+ft_transformer+XGBoost+LightGBM,0.6268590959555621,0.7470252829879999,0.9501275669336527,daejeon,best
129
- deepgbm+resnet_like+XGBoost+LightGBM,0.617990326312608,0.7412687534166494,0.9484927971987257,daejeon,best
130
- ft_transformer+resnet_like+XGBoost+LightGBM,0.5778772266115083,0.7046854387067157,0.9426776372150277,daejeon,best
131
- deepgbm+ft_transformer+resnet_like+XGBoost+LightGBM,0.6251709503819354,0.7454103687111592,0.9500901389824588,daejeon,best
132
- deepgbm+ft_transformer,0.6798477338109503,0.7672365450318365,0.9541375560379602,gwangju,best
133
- deepgbm+resnet_like,0.665680366173557,0.7714595133764687,0.9543727976423164,gwangju,best
134
- deepgbm+XGBoost,0.6574065703012586,0.7719143895834364,0.9543191075928837,gwangju,best
135
- deepgbm+LightGBM,0.6615432493953055,0.7728007686314036,0.9544770483485955,gwangju,best
136
- ft_transformer+resnet_like,0.5874777029243061,0.7627495358829616,0.9528777415974249,gwangju,best
137
- ft_transformer+XGBoost,0.5819133878684225,0.7549252510472145,0.9515179410587,gwangju,best
138
- ft_transformer+LightGBM,0.5805334443959108,0.7489469355258954,0.9505456293509993,gwangju,best
139
- resnet_like+XGBoost,0.5621088257073251,0.742686095459662,0.9492615719369842,gwangju,best
140
- resnet_like+LightGBM,0.5552708987061287,0.7370427725250878,0.9481963158420041,gwangju,best
141
- XGBoost+LightGBM,0.5288440675809284,0.7303546927519567,0.9467888046570956,gwangju,best
142
- deepgbm+ft_transformer+resnet_like,0.6605998838783642,0.7743701608802702,0.9553335870116691,gwangju,best
143
- deepgbm+ft_transformer+XGBoost,0.6495239084733958,0.7651993867637442,0.9536631567565769,gwangju,best
144
- deepgbm+ft_transformer+LightGBM,0.6556854096497969,0.7699591127202252,0.9550316682053713,gwangju,best
145
- deepgbm+resnet_like+XGBoost,0.646377396487214,0.7644362437422076,0.9526370070946761,gwangju,best
146
- deepgbm+resnet_like+LightGBM,0.6441717942778256,0.7623630196443353,0.9527517861450042,gwangju,best
147
- deepgbm+XGBoost+LightGBM,0.6104936580055047,0.7355161749128437,0.9470099225657277,gwangju,best
148
- ft_transformer+resnet_like+XGBoost,0.5926724219238438,0.7169506517948077,0.9453380367792,gwangju,best
149
- ft_transformer+resnet_like+LightGBM,0.5868504053718506,0.7120674745305696,0.9448050003742795,gwangju,best
150
- ft_transformer+XGBoost+LightGBM,0.5697749480416849,0.6978540261423789,0.9408133093794446,gwangju,best
151
- resnet_like+XGBoost+LightGBM,0.5568421888456356,0.6879639430352676,0.9383421289018639,gwangju,best
152
- deepgbm+ft_transformer+resnet_like+XGBoost,0.6464484806055241,0.7628617034260374,0.953282847185834,gwangju,best
153
- deepgbm+ft_transformer+resnet_like+LightGBM,0.6432761028976647,0.7603027833589299,0.9530166928662326,gwangju,best
154
- deepgbm+ft_transformer+XGBoost+LightGBM,0.6268590959555621,0.7470252829879999,0.9501275669336527,gwangju,best
155
- deepgbm+resnet_like+XGBoost+LightGBM,0.617990326312608,0.7412687534166494,0.9484927971987257,gwangju,best
156
- ft_transformer+resnet_like+XGBoost+LightGBM,0.5778772266115083,0.7046854387067157,0.9426776372150277,gwangju,best
157
- deepgbm+ft_transformer+resnet_like+XGBoost+LightGBM,0.6251709503819354,0.7454103687111592,0.9500901389824588,gwangju,best
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/model_result/deepgbm_sampled_data_test.csv DELETED
@@ -1,31 +0,0 @@
1
- region,model,data_sample,CSI,MCC,Accuracy
2
- seoul,deepgbm,pure,0.6442509203188513,0.7660970126461858,0.9575257213197927
3
- busan,deepgbm,pure,0.6818642050717673,0.8049988566475562,0.9749078856534504
4
- incheon,deepgbm,pure,0.5671286321477068,0.6887072986816571,0.9090567324566875
5
- daegu,deepgbm,pure,0.5768403304029271,0.7158561147888857,0.9752356921259908
6
- daejeon,deepgbm,pure,0.6879666308365847,0.797628286801526,0.9573174763580109
7
- gwangju,deepgbm,pure,0.6169039999207967,0.7404995782864335,0.9513605060259002
8
- seoul,deepgbm,smote,0.6178386591753761,0.7437507767516012,0.9473890885046287
9
- busan,deepgbm,smote,0.5771081499427251,0.7328290895553372,0.9590890660478578
10
- incheon,deepgbm,smote,0.559840116480077,0.6829120896700739,0.9007683126647871
11
- daegu,deepgbm,smote,0.6425770288404345,0.7686582033950807,0.9787012085069575
12
- daejeon,deepgbm,smote,0.5569398315371836,0.6906810572279379,0.9299289492726501
13
- gwangju,deepgbm,smote,0.5526168770566259,0.6910361569617187,0.9341257662333341
14
- seoul,deepgbm,ctgan20000,0.7095252362606882,0.8058537207489471,0.9619258968152972
15
- busan,deepgbm,ctgan20000,0.5799669524449778,0.7334430424969706,0.9640024203408438
16
- incheon,deepgbm,ctgan20000,0.5289538894298621,0.6580659191964314,0.8935004283421081
17
- daegu,deepgbm,ctgan20000,0.44609227542135127,0.616365605274888,0.9682904159492977
18
- daejeon,deepgbm,ctgan20000,0.5716353901847903,0.7095774395951602,0.9385713751029269
19
- gwangju,deepgbm,ctgan20000,0.4571493095973662,0.620850435638757,0.9200951709625637
20
- seoul,deepgbm,ctgan10000,0.5616336216829804,0.700657840332931,0.9383062604486363
21
- busan,deepgbm,ctgan10000,0.5879352056252677,0.7351657312694786,0.9672767422711281
22
- incheon,deepgbm,ctgan10000,0.5290269463089048,0.6582877751557777,0.8928201753291581
23
- daegu,deepgbm,ctgan10000,0.5557972102197738,0.7032214449405951,0.9747075421480318
24
- daejeon,deepgbm,ctgan10000,0.5413249801262358,0.6761149432485588,0.9286541116683718
25
- gwangju,deepgbm,ctgan10000,0.41816087724732665,0.5735307327535618,0.9039835816054095
26
- seoul,deepgbm,ctgan7000,0.5813133937275518,0.7156872321691327,0.9403897497317663
27
- busan,deepgbm,ctgan7000,0.597435747844303,0.7454746315945583,0.9670328367891807
28
- incheon,deepgbm,ctgan7000,0.5123361043389418,0.646967905975133,0.8907633014779882
29
- daegu,deepgbm,ctgan7000,0.6146700981464019,0.7399778687018386,0.9766398640949504
30
- daejeon,deepgbm,ctgan7000,0.5581763081057168,0.6896808823936648,0.9308861691244354
31
- gwangju,deepgbm,ctgan7000,0.5798647223742114,0.7012366964380711,0.9421613394216134
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/model_result/ft_transformer_sampled_data_test.csv DELETED
@@ -1,31 +0,0 @@
1
- region,model,data_sample,CSI,MCC,Accuracy
2
- seoul,ft_transformer,pure,0.5998006360910436,0.7351618125239984,0.9481156066239157
3
- busan,ft_transformer,pure,0.525891811597548,0.6847515269647338,0.9593201836464805
4
- incheon,ft_transformer,pure,0.574439584484584,0.6997555499382345,0.9142701341584117
5
- daegu,ft_transformer,pure,0.5325909397972257,0.6846558598973905,0.9751660345501576
6
- daejeon,ft_transformer,pure,0.5841417763225328,0.710099560862753,0.9446515457743843
7
- gwangju,ft_transformer,pure,0.5123621312403577,0.6524953764059983,0.9400000831732248
8
- seoul,ft_transformer,smote,0.6027209460369666,0.7326561727884195,0.9459808618409561
9
- busan,ft_transformer,smote,0.49442102850607217,0.6656582087002536,0.9448999218171686
10
- incheon,ft_transformer,smote,0.5695903712638072,0.6933882645176865,0.9045313812577455
11
- daegu,ft_transformer,smote,0.5152208156718004,0.680387175933169,0.9707628440252515
12
- daejeon,ft_transformer,smote,0.5604907640952831,0.6919263799254157,0.9305719822674684
13
- gwangju,ft_transformer,smote,0.4852746797180565,0.6329231459699102,0.9290042709450974
14
- seoul,ft_transformer,ctgan20000,0.5803899707808177,0.7130066787294806,0.9414579018722292
15
- busan,ft_transformer,ctgan20000,0.4639353486936595,0.6467071571334145,0.9446345992298159
16
- incheon,ft_transformer,ctgan20000,0.5640202503368519,0.68857629970368,0.9072701715863629
17
- daegu,ft_transformer,ctgan20000,0.3506001798809408,0.5079853139279686,0.9495035598140248
18
- daejeon,ft_transformer,ctgan20000,0.5239938645095054,0.669429458663152,0.9276992830468016
19
- gwangju,ft_transformer,ctgan20000,0.4145097694671273,0.5668829126361233,0.9129590122347814
20
- seoul,ft_transformer,ctgan10000,0.5205630914565772,0.6720401513203828,0.9380401061290349
21
- busan,ft_transformer,ctgan10000,0.4161261582289995,0.5928041147082687,0.9366729462451447
22
- incheon,ft_transformer,ctgan10000,0.5623027751442908,0.6844425259764174,0.9051445758581398
23
- daegu,ft_transformer,ctgan10000,0.4662085962650442,0.6230709704268124,0.9675943600236212
24
- daejeon,ft_transformer,ctgan10000,0.5229201335739658,0.6589209425719041,0.9264041719689597
25
- gwangju,ft_transformer,ctgan10000,0.4575447167412667,0.6033826231809732,0.9180866415483528
26
- seoul,ft_transformer,ctgan7000,0.5679676743815468,0.7057250825135037,0.9415375402350475
27
- busan,ft_transformer,ctgan7000,0.4934746936691438,0.6650891221099499,0.9561240445475793
28
- incheon,ft_transformer,ctgan7000,0.5469994030503834,0.6742953373159715,0.9032061198858864
29
- daegu,ft_transformer,ctgan7000,0.37617290206033815,0.5389485667787934,0.9553074914123645
30
- daejeon,ft_transformer,ctgan7000,0.5437623015445493,0.6785419094519339,0.9309658074872537
31
- gwangju,ft_transformer,ctgan7000,0.46978186086856627,0.6149128334271473,0.9244421155941479
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/model_result/lightgbm_sampled_data_test.csv DELETED
@@ -1,31 +0,0 @@
1
- region,model,data_sample,CSI,MCC,Accuracy
2
- seoul,LightGBM,pure,0.4830701326492519,0.6294046921291281,0.9297619790237127
3
- busan,LightGBM,pure,0.43506984201427806,0.603839321192473,0.9573520972128652
4
- incheon,LightGBM,pure,0.5522436617260514,0.6857225827228159,0.9117255533098785
5
- daegu,LightGBM,pure,0.2974860165425261,0.4862244089903067,0.9538572622701301
6
- daejeon,LightGBM,pure,0.48570589458995483,0.6313963695110991,0.9337746712578286
7
- gwangju,LightGBM,pure,0.4841000529921969,0.6363547628117262,0.9430086666500319
8
- seoul,LightGBM,smote,0.5811441298770988,0.7097862880128495,0.9400710923139624
9
- busan,LightGBM,smote,0.468977364244475,0.6345551645621824,0.9506557169116118
10
- incheon,LightGBM,smote,0.5845373158941275,0.7079881355705151,0.9115271851685506
11
- daegu,LightGBM,smote,0.45130042516888275,0.6228321490115053,0.9638063394632
12
- daejeon,LightGBM,smote,0.5294747877274647,0.6632804903422373,0.9322567599038517
13
- gwangju,LightGBM,smote,0.5204263336367001,0.658370785349228,0.9371142841696402
14
- seoul,LightGBM,ctgan20000,0.5278642102510234,0.6732186063208289,0.9409731059377364
15
- busan,LightGBM,ctgan20000,0.4608874696247123,0.6276000553143347,0.958718945197162
16
- incheon,LightGBM,ctgan20000,0.5536461635852592,0.6798438480715302,0.8955189385433041
17
- daegu,LightGBM,ctgan20000,0.4051438319036602,0.5791827944687878,0.965851153196763
18
- daejeon,LightGBM,ctgan20000,0.4682791445662134,0.6207261950649575,0.9317582403872545
19
- gwangju,LightGBM,ctgan20000,0.4817029529408228,0.6339890466307277,0.9420201528723874
20
- seoul,LightGBM,ctgan10000,0.5442974261785484,0.6853617099956927,0.9422277740349827
21
- busan,LightGBM,ctgan10000,0.4518917046571256,0.6185787696267423,0.9571202518485249
22
- incheon,LightGBM,ctgan10000,0.5472322273711279,0.6822157410012658,0.9102826018248206
23
- daegu,LightGBM,ctgan10000,0.42595146846779247,0.5975633254437179,0.9694990268732688
24
- daejeon,LightGBM,ctgan10000,0.48065206351943335,0.6295068759845875,0.9330893238848551
25
- gwangju,LightGBM,ctgan10000,0.48182318531993423,0.6331992271211336,0.9421339962239356
26
- seoul,LightGBM,ctgan7000,0.5395567725547653,0.6810123059399511,0.9410132370187391
27
- busan,LightGBM,ctgan7000,0.4584344058363401,0.6253420379252826,0.9581093894253563
28
- incheon,LightGBM,ctgan7000,0.5498019261194692,0.6849206337149152,0.911005065249395
29
- daegu,LightGBM,ctgan7000,0.39073560033282706,0.5690411701305705,0.9644477089935207
30
- daejeon,LightGBM,ctgan7000,0.4739750751092528,0.6246889480882462,0.9325196912609893
31
- gwangju,LightGBM,ctgan7000,0.48000404186608364,0.6311702463153694,0.9417549342515658
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/model_result/resnet_like_sampled_data_test.csv DELETED
@@ -1,31 +0,0 @@
1
- region,model,data_sample,CSI,MCC,Accuracy
2
- seoul,resnet_like,pure,0.5556513128454227,0.6944843892225018,0.939922004308373
3
- busan,resnet_like,pure,0.5246252526595989,0.6874494677398264,0.9591643378163702
4
- incheon,resnet_like,pure,0.5799566950805518,0.7030239128662202,0.9153711397227005
5
- daegu,resnet_like,pure,0.4394531403147615,0.6143074280063717,0.9659717543728323
6
- daejeon,resnet_like,pure,0.5430465060488938,0.6757025454804458,0.9349603055784282
7
- gwangju,resnet_like,pure,0.5577057247702664,0.689775566381261,0.9437623200339346
8
- seoul,resnet_like,smote,0.6137456175954706,0.7422130871955783,0.9452800234548494
9
- busan,resnet_like,smote,0.5023305919728182,0.6726508273995124,0.9473319069125267
10
- incheon,resnet_like,smote,0.5860766625532056,0.7078408929922378,0.9089022381914814
11
- daegu,resnet_like,smote,0.4901581106489206,0.6522246131616541,0.9662389483577446
12
- daejeon,resnet_like,smote,0.5108150204429799,0.6519519041478157,0.9180820670209847
13
- gwangju,resnet_like,smote,0.50385899296261,0.6510080124725456,0.9282384534770567
14
- seoul,resnet_like,ctgan20000,0.5072327540718837,0.6620612708395112,0.9275254510068119
15
- busan,resnet_like,ctgan20000,0.38986584909236405,0.5796375719673601,0.9392384243664279
16
- incheon,resnet_like,ctgan20000,0.5432016410802436,0.6676977064251691,0.8980009315401186
17
- daegu,resnet_like,ctgan20000,0.2688077976760475,0.440646061562418,0.9229636075554558
18
- daejeon,resnet_like,ctgan20000,0.5006204822135542,0.6492987602101151,0.9202535119894204
19
- gwangju,resnet_like,ctgan20000,0.444672867325423,0.5955785036774024,0.9136339629546457
20
- seoul,resnet_like,ctgan10000,0.5304050574661009,0.6815565974252659,0.9339477755321007
21
- busan,resnet_like,ctgan10000,0.5675115951993178,0.7176500871011381,0.965488621902837
22
- incheon,resnet_like,ctgan10000,0.5608852333036517,0.6835219927895569,0.9037794953048714
23
- daegu,resnet_like,ctgan10000,0.3723832547505761,0.5464472897265699,0.9578420914739127
24
- daejeon,resnet_like,ctgan10000,0.5181167801633592,0.6610880657167911,0.9187646072976188
25
- gwangju,resnet_like,ctgan10000,0.40351880029410764,0.5649610782208994,0.9066332726168792
26
- seoul,resnet_like,ctgan7000,0.5731369888329633,0.7091329811132822,0.9410729138075871
27
- busan,resnet_like,ctgan7000,0.4605071450423024,0.638239831036732,0.9479047624988564
28
- incheon,resnet_like,ctgan7000,0.5343315289869124,0.6644525961066555,0.8994391005647463
29
- daegu,resnet_like,ctgan7000,0.44530071901663565,0.6174979682782439,0.9673712478478929
30
- daejeon,resnet_like,ctgan7000,0.5092618652909245,0.6496900098750791,0.9181710623716013
31
- gwangju,resnet_like,ctgan7000,0.49270227739086486,0.6375371144090152,0.9272533705949381
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/model_result/xgboost_sampled_data_test.csv DELETED
@@ -1,31 +0,0 @@
1
- region,model,data_sample,CSI,MCC,Accuracy
2
- seoul,XGBoost,pure,0.565011886957529,0.7004827765196581,0.9455723773402865
3
- busan,XGBoost,pure,0.4602444563246328,0.6260853753404706,0.9597446789929386
4
- incheon,XGBoost,pure,0.5581443524210103,0.6892177198119791,0.9130578844058522
5
- daegu,XGBoost,pure,0.4166514505499186,0.600956509945712,0.9683242050719033
6
- daejeon,XGBoost,pure,0.5044504047922097,0.6471930212882061,0.9357524265788357
7
- gwangju,XGBoost,pure,0.49003643817143744,0.6379812479286601,0.943428795402184
8
- seoul,XGBoost,smote,0.5833152545530768,0.7120196607674459,0.9418898828089262
9
- busan,XGBoost,smote,0.46473886869778475,0.6265333263141976,0.9518734768903195
10
- incheon,XGBoost,smote,0.5891561059817683,0.7103919570983672,0.9131620588700086
11
- daegu,XGBoost,smote,0.4611201906216012,0.6275336823062336,0.9654021217489666
12
- daejeon,XGBoost,smote,0.5188768068576874,0.6552123484066442,0.9321021616721145
13
- gwangju,XGBoost,smote,0.5099103046808192,0.6499937333212472,0.9358984995550234
14
- seoul,XGBoost,ctgan20000,0.5570923548170014,0.6942604158597989,0.9440149587044938
15
- busan,XGBoost,ctgan20000,0.46709831119818074,0.6290561623391081,0.9589464239671965
16
- incheon,XGBoost,ctgan20000,0.5523124093005438,0.6854497320743039,0.9114981785063753
17
- daegu,XGBoost,ctgan20000,0.4182253863349025,0.5952665980456898,0.9664244246492171
18
- daejeon,XGBoost,ctgan20000,0.4918910489160739,0.6390630711637499,0.933889866174281
19
- gwangju,XGBoost,ctgan20000,0.5013450661414979,0.6465339172273498,0.9428181999650672
20
- seoul,XGBoost,ctgan10000,0.5651854445647394,0.7027118144017459,0.9450405885337393
21
- busan,XGBoost,ctgan10000,0.46692094075656104,0.6325842231097966,0.958222712944249
22
- incheon,XGBoost,ctgan10000,0.5578011155842045,0.6901038363927982,0.9120290316141428
23
- daegu,XGBoost,ctgan10000,0.424541930975018,0.6005603470129561,0.966994369172676
24
- daejeon,XGBoost,ctgan10000,0.4845439338249185,0.63402545260005,0.9338891384085635
25
- gwangju,XGBoost,ctgan10000,0.490651260954804,0.6368315107576099,0.9422102036912277
26
- seoul,XGBoost,ctgan7000,0.5637353305761614,0.7003517728134977,0.9446616305279004
27
- busan,XGBoost,ctgan7000,0.4616459581003174,0.6279091576031567,0.9586027106153988
28
- incheon,XGBoost,ctgan7000,0.5600003713221271,0.6911778795112914,0.9122589016143924
29
- daegu,XGBoost,ctgan7000,0.41141211423242163,0.5889669181000196,0.9661207384118904
30
- daejeon,XGBoost,ctgan7000,0.4880797047418282,0.6352680209224976,0.9336605160066872
31
- gwangju,XGBoost,ctgan7000,0.49570507941388103,0.6441170200639816,0.9430463025342881
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Analysis_code/model_visualize.ipynb DELETED
The diff for this file is too large to render. See raw diff