Debug model_training.py to correct the differences between ensemble a…

…nd predict_labelled predictions
kspruthviraj · Nov 24, 2022 · aee5e7a · aee5e7a
1 parent 9917b15
commit aee5e7a
Showing 1 changed file with 14 additions and 14 deletions.
diff --git a/utils/model_training.py b/utils/model_training.py
@@ -527,13 +527,6 @@ def train_and_save(self, train_main, data_loader):
                     print('If you want to retrain then set "resume from saved" to "yes"')
                     self.run_prediction(data_loader, 'finetuned')
 
-            elif train_main.params.finetune == 0:
-                if not os.path.exists(model_present_path0):
-                    self.train_predict(train_main, data_loader, 0)
-                else:
-                    print('If you want to retrain then set "resume from saved" to "yes"')
-                    self.run_prediction(data_loader, 'original')
-
             elif train_main.params.finetune == 1:
                 if not os.path.exists(model_present_path0):
                     self.train_predict(train_main, data_loader, 0)
@@ -548,6 +541,13 @@ def train_and_save(self, train_main, data_loader):
                     print('If you want to retrain then set "resume from saved" to "yes"')
                     self.run_prediction(data_loader, 'tuned')
 
+            elif train_main.params.finetune == 0:
+                if not os.path.exists(model_present_path0):
+                    self.train_predict(train_main, data_loader, 0)
+                else:
+                    print('If you want to retrain then set "resume from saved" to "yes"')
+                    self.run_prediction(data_loader, 'original')
+
         elif train_main.params.resume_from_saved == 'yes':
             if train_main.params.finetune == 0:
                 if not os.path.exists(model_present_path0):
@@ -1029,21 +1029,21 @@ def load_model_and_run_prediction(self, train_main, test_main, data_loader):
         self.import_deit_models_for_testing(train_main, test_main)
 
         if test_main.params.finetuned == 0:
-            # self.initialize_model(train_main, test_main, data_loader, train_main.params.lr)
+            self.initialize_model(train_main, test_main, data_loader, train_main.params.lr)
             if test_main.params.ensemble == 0:
                 self.run_prediction_on_unseen(test_main, data_loader, 'original')
             else:
                 self.run_ensemble_prediction_on_unseen(test_main, data_loader, 'original')
 
         elif test_main.params.finetuned == 1:
-            # self.initialize_model(train_main, test_main, data_loader, train_main.params.lr)
+            self.initialize_model(train_main, test_main, data_loader, train_main.params.lr)
             if test_main.params.ensemble == 0:
                 self.run_prediction_on_unseen(test_main, data_loader, 'tuned')
             else:
                 self.run_ensemble_prediction_on_unseen(test_main, data_loader, 'tuned')
 
         elif test_main.params.finetuned == 2:
-            # self.initialize_model(train_main, test_main, data_loader, train_main.params.lr)
+            self.initialize_model(train_main, test_main, data_loader, train_main.params.lr)
             if test_main.params.ensemble == 0:
                 self.run_prediction_on_unseen(test_main, data_loader, 'finetuned')
             else:
@@ -1056,21 +1056,21 @@ def load_model_and_run_prediction_with_y(self, train_main, test_main, data_loade
         self.import_deit_models_for_testing(train_main, test_main)
 
         if test_main.params.finetuned == 0:
-            # self.initialize_model(train_main, test_main, data_loader, train_main.params.lr)
+            self.initialize_model(train_main, test_main, data_loader, train_main.params.lr)
             if test_main.params.ensemble == 0:
                 self.run_prediction_on_unseen_with_y(test_main, data_loader, 'original')
             else:
                 self.run_ensemble_prediction_on_unseen_with_y(test_main, data_loader, 'original')
 
         elif test_main.params.finetuned == 1:
-            # self.initialize_model(train_main, test_main, data_loader, train_main.params.lr)
+            self.initialize_model(train_main, test_main, data_loader, train_main.params.lr)
             if test_main.params.ensemble == 0:
                 self.run_prediction_on_unseen_with_y(test_main, data_loader, 'tuned')
             else:
                 self.run_ensemble_prediction_on_unseen_with_y(test_main, data_loader, 'tuned')
 
         elif test_main.params.finetuned == 2:
-            # self.initialize_model(train_main, test_main, data_loader, train_main.params.lr)
+            self.initialize_model(train_main, test_main, data_loader, train_main.params.lr)
             if test_main.params.ensemble == 0:
                 self.run_prediction_on_unseen_with_y(test_main, data_loader, 'finetuned')
             else:
@@ -1367,7 +1367,7 @@ def cls_predict_on_unseen_with_y(val_loader, model, criterion, time_begin=None):
             images, target = images.to(device), target.to(device)
             targets.append(target)
 
-            output, x = model(images)
+            output = model(images)
             outputs.append(output)
             prob = torch.nn.functional.softmax(output, dim=1)
             probs.append(prob)