word vector transformation works for obj atr classification

constants_crunchy.py

@@ -5,7 +5,7 @@ def mkdir_if_not_exists(dir_name):
     if not os.path.exists(dir_name):
         os.mkdir(dir_name)
         
-experiment_name = 'object_attribute_classifier_large_images_vqa_split' #'QA_joint_pretrain_genome_split'
+experiment_name = 'object_attribute_classifier_wordvec_xform' #'QA_joint_pretrain_genome_split'
 
 # Global output directory (all subexperiments will be saved here)
 global_output_dir = '/home/tanmay/Code/GenVQA/Exp_Results/VQA'
@@ -117,7 +117,7 @@ region_model = os.path.join(
     'model')                    
 
 # Object Attribute Finetuning Params
-region_fine_tune_from_iter = 34000
+region_fine_tune_from_iter = 50500
 region_fine_tune_from = region_model + '-' + str(region_fine_tune_from_iter)
 
 # Object Attribute Model Selection
@@ -129,7 +129,7 @@ region_model_accuracies_txt = os.path.join(
 
 # Object Attribute Classifier Evaluation Params
 region_eval_on = 'train_held_out' # One of {'test','train_held_out','train_subset'}
-region_model_to_eval = region_model + '-' + '80000'
+region_model_to_eval = region_model + '-' + '102000'
 
 region_attribute_scores_dirname = os.path.join(
     region_output_dir,

word2vec/word_vector_management.py

@@ -29,16 +29,37 @@ class word_vector_manager():
             
             with tf.variable_scope('attribute_label_word_vectors'):
                 self.attribute_label_word_vectors()
-
-
+    
+    def word_vec_transformation(self, word_vecs):
+        with tf.variable_scope('word_vec_transform'):
+            word_vecs_fc1 = layers.full(
+                word_vecs,
+                constants.word_vector_size,
+                'word_vec_fc1')
+
+            word_vecs_fc2 = layers.full(
+                word_vecs_fc1,
+                constants.word_vector_size,
+                'word_vec_fc2',
+                func = None)
+
+        return word_vecs_fc2
+        
     def init_word_vector_tensor(self):
-#        with tf.device('/cpu:0'):
-        self.word_vectors = tf.get_variable(
-            name = 'word_vector',
-            shape = [self.vocab_size, constants.word_vector_size],
-            initializer = tf.constant_initializer(self.vocab_word_vectors))
+        # self.word_vectors_ = tf.get_variable(
+        #     name = 'word_vector',
+        #     shape = [self.vocab_size, constants.word_vector_size],
+        #     initializer = tf.constant_initializer(self.vocab_word_vectors))
+        
+        self.pretrained_word_vectors = tf.constant(
+            self.vocab_word_vectors,
+            name = 'pretrained_word_vectors'
+        )
+
+        self.word_vectors = self.word_vec_transformation(
+            self.pretrained_word_vectors)
 
-        tf.add_to_collection('to_regularize', self.word_vectors)
+        #tf.add_to_collection('to_regularize', self.word_vectors)
 
     def read_object_labels(self):
         with open(constants.object_labels_json, 'r') as file:
@@ -88,6 +109,7 @@ class word_vector_manager():
         self.object_label_vectors = tf.concat(
             0, object_label_vector_list)
 
+        
     def attribute_label_word_vectors(self):
         inv_attribute_labels = {v: k for k, v in self.attribute_labels.items()}
         num_attribute_labels = len(inv_attribute_labels)
@@ -101,6 +123,9 @@ class word_vector_manager():
         self.attribute_label_vectors = tf.concat(
             0, attribute_label_vector_list)
         
+    def get_trainable_variables():
+        return var_collect.collect_scope('word_vec_transform')
+
 if __name__=='__main__':
     word_vector_mgr = word_vector_manager()