generate dataset splits

.gitattributes

@@ -53,3 +53,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.jpg filter=lfs diff=lfs merge=lfs -text
 *.jpeg filter=lfs diff=lfs merge=lfs -text
 *.webp filter=lfs diff=lfs merge=lfs -text
+train.jsonl filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,134 @@
+.DS_Store
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+pip-wheel-metadata/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+.python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+htmls/
+dataframes/
+urls/

scripts/00_gen_urls.py

@@ -0,0 +1,37 @@
+import os
+from concurrent.futures import ThreadPoolExecutor
+
+import requests
+
+OUTPUT_DIR = "urls"
+NUM_ROWS = 1800
+MAX_WORKERS = 10
+
+def main():
+    if not os.path.isdir(OUTPUT_DIR):
+        os.mkdir(OUTPUT_DIR)
+    for i, grid_size in enumerate(['4x7', '4x6', '4x5', '4x4', '3x5', '3x4']):
+        for j, difficulty in enumerate(['challenging', 'moderate', 'easy']):
+            file_path = f'{OUTPUT_DIR}/{difficulty}{grid_size}.txt'
+            count = 0
+            if os.path.exists(file_path):
+                with open(file_path) as rr:
+                    count = len(rr.readlines())
+            print(f"PROCESS: {file_path}, {count=}")
+            with open(file_path, 'a') as file:
+                with ThreadPoolExecutor(max_workers=MAX_WORKERS) as executor:
+                    for _ in range(NUM_ROWS - count):
+                        executor.submit(fetch_one, file, i, j)
+
+
+def fetch_one(file, i, j):
+    response = requests.post('https://logic.puzzlebaron.com/init2.php', data={'sg': 6 - i, 'sd': 3 - j})
+    html_content = response.text
+    sfx = html_content.partition('<form method="post" action="play.php?u2=')[2].partition('">')[0]
+    if not sfx.isalnum():
+        print("FAIL", sfx)
+    url = f"https://logic.puzzlebaron.com/play.php?u2={sfx}\n"
+    file.write(url)
+
+
+main()

scripts/01_gen_htmls.py

@@ -0,0 +1,40 @@
+import os
+from concurrent.futures import ThreadPoolExecutor
+
+import requests
+
+OUTPUT_DIR = "htmls"
+MAX_WORKERS = 10
+
+def download_url(url, full_path):
+    try:
+        response = requests.get(url)
+        response.raise_for_status()
+        with open(full_path, 'wb') as f:
+            f.write(response.content)
+    except Exception as e:
+        print(f"Failed to download {url}: {e}")
+
+
+def main():
+    if not os.path.isdir(OUTPUT_DIR):
+        os.mkdir(OUTPUT_DIR)
+    for grid_size in ['4x7', '4x6', '4x5', '4x4', '3x5', '3x4']:
+        for difficulty in ['challenging', 'moderate', 'easy']:
+            file_prefix = f"htmls/{difficulty}{grid_size}"
+            if not os.path.isdir(file_prefix):
+                os.mkdir(file_prefix)
+            print(f"PROCESSING: {file_prefix}")
+            with open(f'urls/{difficulty}{grid_size}.txt') as urls:
+                with ThreadPoolExecutor(max_workers=MAX_WORKERS) as executor:
+                    for uu in urls:
+                        executor.submit(process_one_chunk, file_prefix, uu.strip())
+
+def process_one_chunk(file_prefix, url):
+    url_suffix = url.removeprefix("https://logic.puzzlebaron.com/")
+    file_path = os.path.join(file_prefix, url_suffix)
+    if not os.path.exists(file_path):
+        download_url(url, file_path)
+
+
+main()

scripts/02_gen_dataframes.py

@@ -0,0 +1,91 @@
+import os
+import re
+
+import pandas as pd
+from bs4 import BeautifulSoup
+
+
+def extract_puzzle_data(html):
+    """extract data from HTML content"""
+    soup = BeautifulSoup(html, 'html.parser')
+
+    # Extract task description
+    task_description_section = soup.find('div', id='tabs-2') or ""
+    task_description = task_description_section.get_text(strip=True).replace('\xa0', ' ')
+    task_description = task_description.removeprefix("Backstory and Goal")
+    task_description = task_description.partition("Remember, as with all")[0]
+
+    categories = [cc.get_text(strip=True).replace('\xa0', ' ') for cc in soup.find_all('td', class_='answergrid_head')]
+
+    # Extract clues, and clean them
+    clues = []
+    for clue_div in soup.find_all('div', class_='clue'):
+        clue_raw = clue_div.get_text(strip=True).replace('\xa0', ' ')
+        # Remove numbers at the beginning of the string followed by a period and whitespace
+        cleaned_clue = re.sub(r'^\d+\.\s*', '', clue_raw)
+        clues.append(cleaned_clue)
+
+    # Extract label names such as from labelboxh
+    label_categories = dict(label_a=[])
+    for label in soup.find_all('td', class_='labelboxh'):
+        if label['id'].startswith("labelleftA"):
+            label_categories["label_a"].append(label.get_text(strip=True).replace('\xa0', ' '))
+    for letter in "bcd":
+        pattern = re.compile(f'label{letter}_ary' + r'\[\d+]\s*=\s*"([^"]+)";')
+        items = pattern.findall(html)
+        label_categories[f"label_{letter}"] = items
+    return dict(story=task_description,
+                clues=clues,
+                categories=categories,
+                **label_categories)
+
+
+global_stories = set()
+global_clues = set()
+
+
+def process_one(difficulty, grid_size):
+    puzzle_data = []
+    with open(f'urls/{difficulty}{grid_size}.txt') as rr:
+        all_paths = [p.strip() for p in rr]
+    dir_path = f'htmls/{difficulty}{grid_size}/'
+    for c, puzzle_url in enumerate(all_paths):
+        filename = puzzle_url.removeprefix("https://logic.puzzlebaron.com/")
+        if c % 200 == 0:
+            print(f"{c=}")
+        file_path = os.path.join(dir_path, filename)
+        with open(file_path, 'r', encoding='utf-8') as file:
+            html_content = file.read()
+            data = extract_puzzle_data(html_content)
+            if len(global_clues.intersection(data['clues'])) >= 3:
+                continue
+            # elif len(global_clues.intersection(data['clues'])) >= 4:
+            #     print("FAIL:", difficulty, grid_size)
+            #     print(filename)
+            #     print(global_clues.intersection(data['clues']))
+            #     continue
+            #     raise RuntimeError(global_clues.intersection(data['clues']))
+            global_clues.update(data['clues'])
+            data['grid_size'] = grid_size
+            data['difficulty'] = difficulty
+            data['url'] = puzzle_url
+            puzzle_data.append(data)
+    return puzzle_data
+
+
+OUTPUT_DIR = "dataframes"
+
+
+def main():
+    if not os.path.exists(OUTPUT_DIR):
+        os.makedirs(OUTPUT_DIR)
+    for grid_size in ['4x7', '4x6', '4x5', '4x4', '3x5', '3x4']:
+        for difficulty in ['challenging', 'moderate', 'easy']:
+            puzzle_data = process_one(difficulty, grid_size)
+            df = pd.DataFrame(puzzle_data)
+            jsonl_file_path = f'{OUTPUT_DIR}/{difficulty}{grid_size}.jsonl'
+            df.to_json(jsonl_file_path, orient='records', lines=True)
+            print(f'Data saved to {jsonl_file_path}', df.shape)
+
+
+main()

scripts/03_gen_splits.py

@@ -0,0 +1,31 @@
+import os
+from collections import Counter
+
+import pandas as pd
+
+dfs = []
+base_path = "dataframes"
+for file_path in os.listdir(base_path):
+    dfs.append(pd.read_json(os.path.join(base_path, file_path), lines=True))
+raw_df = pd.concat(dfs, ignore_index=True)
+print(f"{raw_df=}")
+
+clue_counts = Counter(clue for clues in raw_df['clues'] for clue in clues)
+
+# take 20% of data with unique stories for a test set
+unique_stories = set(c for c in raw_df['story'])
+test_stories = set(x for i, x in enumerate(unique_stories) if i % 5 == 0)
+test_df0 = raw_df[raw_df['story'].apply(lambda story: story in test_stories)]
+test_df = test_df0[test_df0['clues'].apply(lambda clues: sum(clue_counts[clue] > 1 for clue in clues) <= 1)]
+test_df.to_json("test.jsonl", orient='records', lines=True)
+print(f"{test_df=}")
+
+train_val_df = raw_df[raw_df['story'].apply(lambda story: story not in test_stories)]
+
+val_df = train_val_df[train_val_df.index % 5 == 0]
+val_df.to_json("validation.jsonl", orient='records', lines=True)
+print(f"{val_df=}")
+
+train_df = train_val_df[train_val_df.index % 5 != 0]
+train_df.to_json("train.jsonl", orient='records', lines=True)
+print(f"{train_df=}")

train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:38cd68181ef5fff4709385bee393d74397302af9846df6f34969763f66248109
+size 27153998