fix bugs with training

configs/bengali.yml

@@ -3,7 +3,7 @@ MODEL:
   encodings_len: 256
   mode : 'triplet'
   distance_type : 'l1'
-  backbone : 'resnet18'
+  backbone_name : 'efficientnet-b0'
   backbone_weights : 'imagenet'
   freeze_backbone : False
   embeddings_normalization: True
@@ -12,18 +12,18 @@ DATALOADER:
   dataset_path : '/home/rauf/datasets/bengali/pngs/train/'
   csv_file : '/home/rauf/datasets/bengali/train_new.csv'
   image_id_column : 'image_id'
-  label_column : 'label'
+  label_column : 'grapheme_root'
   validate : True
   val_ratio : 0.2
 
 GENERATOR:
   negatives_selection_mode : 'semihard'
-  mining_n_classes: 5
-  mining_n_samples: 3
+  k_classes: 3
+  k_samples: 5
   margin: 0.5
-  batch_size : 8
-  n_batches : 200
-  augmentation_type : 'default'
+  batch_size : 10
+  n_batches : 10
+  augmentations : 'default'
 
 TRAIN:
   # optimizer parameters
@@ -38,23 +38,21 @@ TRAIN:
   # plot training history
   plot_history : True
 
-SOFTMAX_PRETRAINING:
-  # softmax pretraining parameters
-  optimizer : 'radam'
-  learning_rate : 0.0001
-  decay_factor : 0.99
-  step_size : 1
+# SOFTMAX_PRETRAINING:
+#   # softmax pretraining parameters
+#   optimizer : 'radam'
+#   learning_rate : 0.0001
+#   decay_factor : 0.99
+#   step_size : 1
 
-  batch_size : 8
-  val_steps : 200
-  steps_per_epoch : 500
-  n_epochs : 20
+#   batch_size : 8
+#   val_steps : 200
+#   steps_per_epoch : 10
+#   n_epochs : 1
 
 SAVE_PATHS:
-  work_dir : 'work_dirs/road_signs_resnet18/'
-  encodings_path : 'encodings/'
-  model_save_name : 'best_model_resnet18.h5'
-  encodings_save_name: 'encodings_resnet18.pkl'
+  project_name : 'bengali_efficientnet'
+  work_dir : 'work_dirs/'
 
 ENCODINGS:
   # encodings parameters

embedding_net/backbones.py

@@ -10,7 +10,8 @@ def get_backbone(input_shape,
                  backbone_name='simple',
                  embeddings_normalization=True,
                  backbone_weights='imagenet',
-                 freeze_backbone=False):
+                 freeze_backbone=False,
+                 **kwargs):
     if backbone_name == 'simple':
         input_image = Input(input_shape)
         x = Conv2D(64, (10, 10), activation='relu',

embedding_net/datagenerators.py

@@ -130,9 +130,9 @@ def __init__(self, embedding_model,
                        k_classes=5,
                        k_samples=5,
                        margin=0.5,
-                       negative_selection_mode='semihard'):
+                       negatives_selection_mode='semihard'):
         super().__init__(class_files_paths=class_files_paths, 
-                         clas_names=class_names, 
+                         class_names=class_names, 
                          input_shape=input_shape, 
                          batch_size=batch_size, 
                          n_batches=n_batches, 
@@ -141,10 +141,10 @@ def __init__(self, embedding_model,
                  'hardest': self.hardest_negative,
                  'random_hard': self.random_hard_negative}
         self.embedding_model = embedding_model
-        self.k_classes=k_classes,
-        self.k_samples=k_samples,
-        self.margin=margin
-        self.negative_selection_fn = modes[negative_selection_mode]
+        self.k_classes = k_classes
+        self.k_samples = k_samples
+        self.margin = margin
+        self.negative_selection_fn = modes[negatives_selection_mode]
 
     def hardest_negative(self, loss_values, margin=0.5):
         hard_negative = np.argmax(loss_values)
@@ -160,7 +160,6 @@ def semihard_negative(self, loss_values, margin=0.5):
         return np.random.choice(semihard_negatives) if len(semihard_negatives) > 0 else None
 
     def get_batch_triplets_mining(self):
-
         selected_classes_idxs = np.random.choice(self.n_classes, size=self.k_classes, replace=False)
         selected_classes = [self.class_names[cl] for cl in selected_classes_idxs]
         selected_classes_n_elements = [self.n_samples[cl] for cl in selected_classes]
@@ -228,9 +227,10 @@ def __init__(self, class_files_paths,
                        input_shape=None,
                        batch_size = 32,
                        n_batches = 10, 
-                       augmentations=None):
+                       augmentations=None,
+                       **kwargs):
         super().__init__(class_files_paths=class_files_paths, 
-                         clas_names=class_names, 
+                         class_names=class_names, 
                          input_shape=input_shape, 
                          batch_size=batch_size, 
                          n_batches=n_batches, 
@@ -284,7 +284,7 @@ def __init__(self, class_files_paths,
                        augmentations=None):
 
         super().__init__(class_files_paths=class_files_paths, 
-                         clas_names=class_names, 
+                         class_names=class_names, 
                          input_shape=input_shape, 
                          batch_size=batch_size, 
                          n_batches=n_batches, 
@@ -346,7 +346,7 @@ def __init__(self, class_files_paths,
                        augmentations=None):
 
         super().__init__(class_files_paths=class_files_paths, 
-                         clas_names=class_names, 
+                         class_names=class_names, 
                          input_shape=input_shape, 
                          batch_size=batch_size, 
                          n_batches=n_batches, 

embedding_net/losses_and_accuracies.py

@@ -1,5 +1,4 @@
-import keras.backend as K
-import keras
+import tensorflow.keras.backend as K
 
 
 def contrastive_loss(y_true, y_pred):
@@ -51,30 +50,30 @@ def accuracy(y_true, y_pred):
     return K.mean(K.equal(y_true, K.cast(y_pred < 0.5, y_true.dtype)))
 
 
-class tSNECallback(keras.callbacks.Callback):
+# class tSNECallback(keras.callbacks.Callback):
 
-    def __init__(self, save_file_name='tSNE.gif'):
-        super(tSNECallback, self).__init__()
-        self.save_file_name = save_file_name
+#     def __init__(self, save_file_name='tSNE.gif'):
+#         super(tSNECallback, self).__init__()
+#         self.save_file_name = save_file_name
 
-    def on_train_begin(self, logs={}):
-        self.aucs = []
-        self.losses = []
+#     def on_train_begin(self, logs={}):
+#         self.aucs = []
+#         self.losses = []
 
-    def on_train_end(self, logs={}):
-        return
+#     def on_train_end(self, logs={}):
+#         return
 
-    def on_epoch_begin(self, epoch, logs={}):
-        return
+#     def on_epoch_begin(self, epoch, logs={}):
+#         return
 
-    def on_epoch_end(self, epoch, logs={}):
-        self.losses.append(logs.get('loss'))
-        y_pred = self.model.predict(self.model.validation_data[0])
-        self.aucs.append(roc_auc_score(self.model.validation_data[1], y_pred))
-        return
+#     def on_epoch_end(self, epoch, logs={}):
+#         self.losses.append(logs.get('loss'))
+#         y_pred = self.model.predict(self.model.validation_data[0])
+#         self.aucs.append(roc_auc_score(self.model.validation_data[1], y_pred))
+#         return
 
-    def on_batch_begin(self, batch, logs={}):
-        return
+#     def on_batch_begin(self, batch, logs={}):
+#         return
 
-    def on_batch_end(self, batch, logs={}):
-        return
+#     def on_batch_end(self, batch, logs={}):
+#         return