Changes for submission-StrategyA-max10epoch

local_evaluation.py

@@ -42,7 +42,7 @@ training_dataset = ZEWDPCBaseDataset(
 unlabelled_dataset = ZEWDPCProtectedDataset(
     images_dir="./data/unlabelled/images",
     labels_path="./data/unlabelled/labels.csv",
-    purchase_budget=500,  # Configurable Parameter
+    purchase_budget=1899,  # Configurable Parameter
     shuffle_seed=DATASET_SHUFFLE_SEED,
 )
 # Instantiate Validation Dataset
@@ -75,8 +75,8 @@ checkpoint_folder_path = tempfile.TemporaryDirectory().name
 ####################################################################################
 time_started = time.time()
 
-PURCHASE_BUDGET = 500
-COMPUTE_BUDGET = 900
+PURCHASE_BUDGET = 1899
+COMPUTE_BUDGET = 3060
 
 # ####################################################################################
 ####################################################################################

run.py

@@ -209,7 +209,7 @@ class ZEWDPCBaseRun:
 
     @staticmethod
     def generate_valid_epoch_comb(num_label_list):
-        pos_epoch = [5, 7, 10, 12, 15]
+        pos_epoch = [5, 7, 10]
         return list(combinations_with_replacement(pos_epoch, len(num_label_list)))[::-1]
 
     def calculate_total_time_given_epochs(self, epochs, num_labels_list):
@@ -277,9 +277,9 @@ class ZEWDPCBaseRun:
         print(f"labels: {num_labels_to_purchase}, epochs: {num_epochs_to_train}")
         for e, (num_labels, num_epochs) in enumerate(zip(num_labels_to_purchase, num_epochs_to_train)):
             if e == len(num_labels_to_purchase) - 1:
-                """this is the last training, maximize number of epochs here!"""
+                #this is the last training, maximize number of epochs here!
                 epochs_affordable = int((remaining_time - self.safety_time - self.prediction_time) / (self.train_time_per_image * len(image_lst)))
-                num_epochs = min(20, max(epochs_affordable, num_epochs))
+                num_epochs = min(10, max(epochs_affordable, num_epochs))
                 print(f"last training epoch is {num_epochs} instead!")
             """start training"""
             #self.model = load_model()