new bot 2

crypto/gogo/model/transformer.py

@@ -133,6 +133,7 @@ class Transformer(nn.Module):
     def __init__(self, input_dim, d_model, num_heads, num_layers, d_ff, dropout=0.1):
         super(Transformer, self).__init__()
 
+        self.input_dim = input_dim
         self.candle_embedding = nn.Linear(input_dim, d_model)
         self.tick_embedding = nn.Linear(2, d_model)  # Each tick has price and quantity
 
@@ -152,10 +153,11 @@ class Transformer(nn.Module):
         self.future_ticks_decoder = Decoder(num_layers, d_model, num_heads, d_ff, dropout)
         self.future_ticks_projection = nn.Linear(d_model, 60)  # 30 ticks * (price, quantity) = 60
 
-    def forward(self, candle_data, tick_data, future_candle_mask, future_ticks_mask):
-        # candle_data: [batch_size, seq_len, input_dim]
-        # tick_data:   [batch_size, tick_seq_len, 2]
-
+    def forward(self, candle_data, tick_data, future_candle_mask=None, future_ticks_mask=None):
+        # Print shapes for debugging
+        # print(f"Candle data shape: {candle_data.shape}, Expected input dim: {self.input_dim}")
+        
+        # Embed candle data
         candle_embedded = self.candle_embedding(candle_data)
         candle_embedded = self.positional_encoding(candle_embedded)  # Add positional info
 
@@ -189,7 +191,7 @@ class Transformer(nn.Module):
 
 # Example instantiation (adjust parameters for ~1B parameters)
 if __name__ == '__main__':
-    input_dim = 6 + len([5, 10, 20, 60, 120, 200]) # OHLCV + EMAs
+    input_dim = 11  # Changed from 12 to 11 to match your data
     d_model = 512  # Hidden dimension
     num_heads = 8
     num_layers = 6  # Number of encoder/decoder layers
@@ -220,3 +222,22 @@ if __name__ == '__main__':
     print("Future Candle Prediction Shape:", future_candle_pred.shape)  # Expected: [batch_size, 1, 5]
     print("Future Volume Prediction Shape:", future_volume_pred.shape)  # Expected: [batch_size, 1, 1]
     print("Future Ticks Prediction Shape:", future_ticks_pred.shape)   # Expected: [batch_size, 30, 2]
+
+# Make sure to use this when instantiating the model
+def create_model(input_dim=11):
+    d_model = 512  # Hidden dimension
+    num_heads = 8
+    num_layers = 6  # Number of encoder/decoder layers
+    d_ff = 2048  # Feedforward dimension
+    dropout = 0.1
+    
+    model = Transformer(
+        input_dim=input_dim,
+        d_model=d_model,
+        num_heads=num_heads,
+        num_layers=num_layers,
+        d_ff=d_ff,
+        dropout=dropout
+    )
+    
+    return model