fix f1 score

Author: iyuge2

config/config_regression.py

@@ -96,7 +96,7 @@ def __SELF_MM(self):
                 'need_normalized': False,
                 'use_bert': True,
                 'use_finetune': True,
-                'save_labels': True,
+                'save_labels': False,
                 'early_stop': 8,
                 'update_epochs': 4
             },
@@ -174,12 +174,12 @@ def __SELF_MM(self):
                     # the batch_size of each epoch is update_epochs * batch_size
                     'batch_size': 32,
                     'learning_rate_bert': 5e-5,
-                    'learning_rate_audio': 0.005,
-                    'learning_rate_video': 0.001,
-                    'learning_rate_other': 0.001,
-                    'weight_decay_bert': 0.01,
-                    'weight_decay_audio': 0.0,
-                    'weight_decay_video': 0.0,
+                    'learning_rate_audio': 5e-3,
+                    'learning_rate_video': 5e-3,
+                    'learning_rate_other': 1e-3,
+                    'weight_decay_bert': 0.001,
+                    'weight_decay_audio': 0.01,
+                    'weight_decay_video': 0.01,
                     'weight_decay_other': 0.001,
                     # feature subNets
                     'a_lstm_hidden_size': 16,
@@ -195,12 +195,12 @@ def __SELF_MM(self):
                     # post feature
                     'post_fusion_dim': 128,
                     'post_text_dim':64,
-                    'post_audio_dim': 32,
-                    'post_video_dim': 16,
-                    'post_fusion_dropout': 0.1,
+                    'post_audio_dim': 16,
+                    'post_video_dim': 32,
+                    'post_fusion_dropout': 0.0,
                     'post_text_dropout': 0.1,
-                    'post_audio_dropout': 0.0,
-                    'post_video_dropout': 0.1,
+                    'post_audio_dropout': 0.1,
+                    'post_video_dropout': 0.0,
                     # res
                     'H': 1.0
                 },

utils/metricsTop.py

@@ -35,7 +35,7 @@ def __eval_mosi_classification(self, y_pred, y_true):
         # three classes
         y_pred_3 = np.argmax(y_pred, axis=1)
         Mult_acc_3 = accuracy_score(y_pred_3, y_true)
-        F1_score_3 = f1_score(y_pred_3, y_true, average='weighted')
+        F1_score_3 = f1_score(y_true, y_pred_3, average='weighted')
         # two classes 
         y_pred = np.array([[v[0], v[2]] for v in y_pred])
         # with 0 (<= 0 or > 0)
@@ -45,14 +45,14 @@ def __eval_mosi_classification(self, y_pred, y_true):
             y_true_2.append(0 if v <= 1 else 1)
         y_true_2 = np.array(y_true_2)
         Has0_acc_2 = accuracy_score(y_pred_2, y_true_2)
-        Has0_F1_score = f1_score(y_pred_2, y_true_2, average='weighted')
+        Has0_F1_score = f1_score(y_true_2, y_pred_2, average='weighted')
         # without 0 (< 0 or > 0)
         non_zeros = np.array([i for i, e in enumerate(y_true) if e != 1])
         y_pred_2 = y_pred[non_zeros]
         y_pred_2 = np.argmax(y_pred_2, axis=1)
         y_true_2 = y_true[non_zeros]
         Non0_acc_2 = accuracy_score(y_pred_2, y_true_2)
-        Non0_F1_score = f1_score(y_pred_2, y_true_2, average='weighted')
+        Non0_F1_score = f1_score(y_true_2, y_pred_2, average='weighted')
 
         eval_results = {
             "Has0_acc_2":  round(Has0_acc_2, 4),
@@ -103,12 +103,12 @@ def __eval_mosei_regression(self, y_pred, y_true, exclude_zero=False):
         non_zeros_binary_preds = (test_preds[non_zeros] > 0)
 
         non_zeros_acc2 = accuracy_score(non_zeros_binary_preds, non_zeros_binary_truth)
-        non_zeros_f1_score = f1_score(non_zeros_binary_preds, non_zeros_binary_truth, average='weighted')
+        non_zeros_f1_score = f1_score(non_zeros_binary_truth, non_zeros_binary_preds, average='weighted')
 
         binary_truth = (test_truth >= 0)
         binary_preds = (test_preds >= 0)
         acc2 = accuracy_score(binary_preds, binary_truth)
-        f_score = f1_score(binary_preds, binary_truth, average='weighted')
+        f_score = f1_score(binary_truth, binary_preds, average='weighted')
 
         eval_results = {
             "Has0_acc_2":  round(acc2, 4),
@@ -164,7 +164,7 @@ def __eval_sims_regression(self, y_pred, y_true):
         mult_a2 = self.__multiclass_acc(test_preds_a2, test_truth_a2)
         mult_a3 = self.__multiclass_acc(test_preds_a3, test_truth_a3)
         mult_a5 = self.__multiclass_acc(test_preds_a5, test_truth_a5)
-        f_score = f1_score(test_preds_a2, test_truth_a2, average='weighted')
+        f_score = f1_score(test_truth_a2, test_preds_a2, average='weighted')
 
         eval_results = {
             "Mult_acc_2": mult_a2,