Upload folder using huggingface_hub

README.md

@@ -30,7 +30,7 @@ Input: (batch, 100, 40) - 100 consecutive LOB snapshots
   ↓ 3-class Classification
 ```
 
-Parameters: 85,803
+Parameters: 338,395
 
 ## 输出 / Output Classes
 
@@ -44,13 +44,13 @@ Parameters: 85,803
 
 | Metric | Value |
 |--------|-------|
-| Test Accuracy | 0.1579 |
-| Test F1 (Macro) | 0.1634 |
-| Test F1 (Weighted) | 0.0725 |
-| 主力买入 Precision | 0.1306 |
-| 主力买入 Recall | 0.4739 |
-| 主力卖出 Precision | 0.1876 |
-| 主力卖出 Recall | 0.5947 |
+| Test Accuracy | 0.3980 |
+| Test F1 (Macro) | 0.3795 |
+| Test F1 (Weighted) | 0.4203 |
+| 主力买入 Precision | 0.2418 |
+| 主力买入 Recall | 0.5401 |
+| 主力卖出 Precision | 0.2448 |
+| 主力卖出 Recall | 0.6246 |
 
 ## 使用方法 / Usage
 

config.json

@@ -7,7 +7,7 @@
   "d_model": 64,
   "nhead": 4,
   "dropout": 0.4,
-  "total_parameters": 85803,
+  "total_parameters": 338395,
   "class_names": [
     "主力买入 (Buy)",
     "中性 (Neutral)",
@@ -18,18 +18,18 @@
     "中性/散户",
     "主力卖出"
   ],
-  "test_accuracy": 0.15789473684210525,
-  "test_f1_macro": 0.16335941375062,
-  "test_f1_weighted": 0.07250430112144952,
+  "test_accuracy": 0.3979949874686717,
+  "test_f1_macro": 0.37945545254276664,
+  "test_f1_weighted": 0.4203155400216087,
   "test_precision": [
-    0.13064361191162344,
-    0.0,
-    0.18763102725366876
+    0.24180967238689546,
+    0.7696245733788396,
+    0.24479166666666666
   ],
   "test_recall": [
-    0.4738675958188153,
-    0.0,
-    0.5946843853820598
+    0.5400696864111498,
+    0.3205401563610519,
+    0.6245847176079734
   ],
   "training_dataset": "LeonardoBerti/TRADES-LOB",
   "normalization": "z-score (means/stds in norm_stats.npz)",

model.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b8cba3876b1f6e97f0a5c424e4313e38fd8a83e5f5cb5550d2a0d55bd1d56feb
-size 366176
+oid sha256:af087d34c5d65978edcad41f0dd607b51b6a937a55de3ac73a0f194ed0029b91
+size 1377888

model.py

@@ -16,24 +16,24 @@ class BilinearNorm(nn.Module):
         return gate * (self.gamma * x_norm + self.beta) + (1 - gate) * x
 
 class LOBPatternNetV3(nn.Module):
-    def __init__(self, num_classes=3, d_model=64, nhead=4, dropout=0.4):
+    def __init__(self, num_classes=3, d_model=128, nhead=4, dropout=0.25):
         super().__init__()
         self.norm = BilinearNorm(40)
         self.spatial = nn.Sequential(
-            nn.Conv2d(1, 16, kernel_size=(1, 2), stride=(1, 2)),
-            nn.BatchNorm2d(16), nn.ReLU(), nn.Dropout2d(dropout * 0.5),
-            nn.Conv2d(16, 16, kernel_size=(1, 2), stride=(1, 2)),
-            nn.BatchNorm2d(16), nn.ReLU(), nn.Dropout2d(dropout * 0.5),
-            nn.Conv2d(16, 16, kernel_size=(1, 10)),
-            nn.BatchNorm2d(16), nn.ReLU(),
+            nn.Conv2d(1, 32, kernel_size=(1, 2), stride=(1, 2)),
+            nn.BatchNorm2d(32), nn.LeakyReLU(0.01),
+            nn.Conv2d(32, 32, kernel_size=(1, 2), stride=(1, 2)),
+            nn.BatchNorm2d(32), nn.LeakyReLU(0.01),
+            nn.Conv2d(32, 32, kernel_size=(1, 10)),
+            nn.BatchNorm2d(32), nn.LeakyReLU(0.01),
         )
         self.temporal = nn.Sequential(
-            nn.Conv1d(16, 32, kernel_size=3, padding=1),
-            nn.BatchNorm1d(32), nn.ReLU(), nn.Dropout(dropout),
-            nn.Conv1d(32, 32, kernel_size=5, padding=2),
-            nn.BatchNorm1d(32), nn.ReLU(), nn.Dropout(dropout),
-            nn.Conv1d(32, d_model, kernel_size=3, padding=1),
-            nn.BatchNorm1d(d_model), nn.ReLU(), nn.Dropout(dropout),
+            nn.Conv1d(32, 64, kernel_size=3, padding=1),
+            nn.BatchNorm1d(64), nn.LeakyReLU(0.01), nn.Dropout(dropout),
+            nn.Conv1d(64, 64, kernel_size=5, padding=2),
+            nn.BatchNorm1d(64), nn.LeakyReLU(0.01), nn.Dropout(dropout),
+            nn.Conv1d(64, d_model, kernel_size=3, padding=1),
+            nn.BatchNorm1d(d_model), nn.LeakyReLU(0.01), nn.Dropout(dropout),
         )
         encoder_layer = nn.TransformerEncoderLayer(
             d_model=d_model, nhead=nhead, dim_feedforward=d_model*2,
@@ -43,10 +43,10 @@ class LOBPatternNetV3(nn.Module):
         self.classifier = nn.Sequential(
             nn.LayerNorm(d_model),
             nn.Dropout(dropout),
-            nn.Linear(d_model, 32),
+            nn.Linear(d_model, 64),
             nn.GELU(),
             nn.Dropout(dropout),
-            nn.Linear(32, num_classes)
+            nn.Linear(64, num_classes)
         )
     def forward(self, x):
         x = self.norm(x)