train_loop

Util_funs.py

@@ -242,22 +242,20 @@ def train_loop(data_train_loader, data_test_loader, model, device, epoch = 4, lr
         input_ids, attention_mask,q_token_type_ids, label_id = batch
 
         # Predictions
-        _, feature, prediction = model_meta(input_ids, attention_mask,q_token_type_ids, labels = label_id.squeeze())
+          _, feature, _ = model_meta(input_ids, attention_mask,q_token_type_ids, labels = label_id.squeeze())
 
-        # Save batch's predictions 
-        prediction = prediction.detach().cpu().squeeze()
-        label_id = label_id.detach().cpu()
-        labels.append(label_id.numpy().squeeze())
-        
+ #       prediction = prediction.detach().cpu().squeeze()
+  #      label_id = label_id.detach().cpu()
         logit = feature[1].detach().cpu()
-        predi_logit.append(logit.numpy())
+   #     feature_lat = feature[0].detach().cpu()
 
-        feature_lat = feature[0].detach().cpu()
-        features.append(feature_lat.numpy())
+#        labels.append(label_id.numpy().squeeze())
+#        features.append(feature_lat.numpy())
+        predi_logit.append(logit.numpy())
 
         # Accuracy over the test's bach
-        acc = fn.accuracy(prediction, label_id).item()
-        all_acc.append(acc)
+#        acc = fn.accuracy(prediction, label_id).item()
+ #       all_acc.append(acc)
       del input_ids, attention_mask, label_id, batch
 
   if print_info:
@@ -268,9 +266,13 @@ def train_loop(data_train_loader, data_test_loader, model, device, epoch = 4, lr
   torch.cuda.empty_cache()
 
   del model_meta, optimizer
+  
+  logits = np.concatenate(np.array(predi_logit,dtype=object))
+  logits = torch.tensor(logits.astype(np.float32)).detach().clone()
+#  return features, labels, predi_logit
 
-  return map_feature_tsne(features, labels, predi_logit)
-
+  return logits.detach().clone()
+  
 # Process predictions and map the feature_map in tsne
 def map_feature_tsne(features, labels, predi_logit):