fix: remove hardcoded class weight for debug.

freqtrade/freqai/prediction_models/PyTorchTransformerClassifier.py

@@ -51,9 +51,7 @@ class PyTorchTransformerClassifier(BasePyTorchClassifier):
         )
         model.to(self.device)
         optimizer = torch.optim.AdamW(model.parameters(), lr=self.learning_rate)
-        criterion = torch.nn.CrossEntropyLoss(
-            weight=torch.tensor([10.0, 10.0, 1.0]).to(self.device)
-        )
+        criterion = torch.nn.CrossEntropyLoss()
         # check if continual_learning is activated, and retrieve the model to continue training
         trainer = self.get_init_model(dk.pair)
         if trainer is None: