detach tensor to avoid warnings

Author: florian-huber

ms2deepscore/models/EmbeddingEvaluatorModel.py

@@ -134,7 +134,8 @@ def train_evaluator(
 
                     # Calculate loss
                     loss = criterion(outputs.to(device), mse_per_embedding.to(device, dtype=float32))
-                    iteration_losses.append(float(loss))
+                    loss_value = loss.detach().item()
+                    iteration_losses.append(loss_value)
 
                     # Backward pass and optimize
                     loss.backward()
@@ -159,7 +160,7 @@ def train_evaluator(
                                 mse_per_embedding = mse_per_embedding.reshape(-1, 1).clone().detach()
 
                                 loss = criterion(outputs.to(device), mse_per_embedding.to(device, dtype=float32))
-                                val_losses.append(float(loss))
+                                val_losses.append(loss_value)
                             print(f">>> Val_loss: {np.mean(val_losses):.6f}")
 
                         self.train()

ms2deepscore/models/SiameseSpectralModel.py

@@ -276,7 +276,8 @@ def train(
                 loss = criterion(outputs, targets.to(device), weighting_factor=weighting_factor)
                 if lambda_l1 > 0 or lambda_l2 > 0:
                     loss += l1_regularization(model, lambda_l1) + l2_regularization(model, lambda_l2)
-                batch_losses.append(float(loss))
+                loss_value = loss.detach().item()
+                batch_losses.append(loss_value)
 
                 if monitor_rmse:
                     batch_rmse.append(rmse_loss(outputs, targets.to(device)).cpu().detach().numpy())
@@ -287,7 +288,7 @@ def train(
 
                 # Print progress
                 training.set_postfix(
-                    loss=float(loss),
+                    loss=loss_value,
                     rmse=np.mean(batch_rmse),
                 )
         epoch_loss = np.mean(batch_losses)