Update dien.py

add a global variable initializer op in model def

deepctr/models/dien.py

@@ -116,7 +116,7 @@ def interest_evolution(concat_behavior, deep_input_item, user_behavior_length, g
         hist = AttentionSequencePoolingLayer(hidden_size=att_hidden_size, activation=att_activation, weight_normalization=att_weight_normalization, return_score=False)([
             deep_input_item, rnn_outputs2, user_behavior_length])
 
-    else:#AIGRU AGRU AUGRU
+    else:  # AIGRU AGRU AUGRU
 
         scores = AttentionSequencePoolingLayer(hidden_size=att_hidden_size, activation=att_activation, weight_normalization=att_weight_normalization, return_score=True)([
             deep_input_item, rnn_outputs, user_behavior_length])
@@ -125,7 +125,7 @@ def interest_evolution(concat_behavior, deep_input_item, user_behavior_length, g
             hist = multiply([rnn_outputs, Permute([2, 1])(scores)])
             final_state2 = DynamicGRU(embedding_size * 2, gru_type="GRU", return_sequence=False, name='gru2')(
                 [hist, user_behavior_length])
-        else:#AGRU AUGRU
+        else:  # AGRU AUGRU
             final_state2 = DynamicGRU(embedding_size * 2, gru_type=gru_type, return_sequence=False,
                                       name='gru2')([rnn_outputs, user_behavior_length, Permute([2, 1])(scores)])
         hist = final_state2
@@ -220,4 +220,5 @@ def DIEN(feature_dim_dict, seq_feature_list, embedding_size=8, hist_len_max=16,
 
     if use_negsampling:
         model.add_loss(alpha * aux_loss_1)
+    tf.keras.backend.get_session().run(tf.global_variables_initializer())
     return model