fix drop rate in attention_lstm.py

fluid/PaddleCV/video/models/attention_lstm/attention_lstm.py

@@ -34,6 +34,7 @@ class AttentionLSTM(ModelBase):
         self.num_classes = self.cfg.MODEL.num_classes
         self.embedding_size = self.cfg.MODEL.embedding_size
         self.lstm_size = self.cfg.MODEL.lstm_size
+        self.drop_rate = self.cfg.MODEL.drop_rate
 
         # get mode configs
         self.batch_size = self.get_config_from_sec(self.mode, 'batch_size', 1)
@@ -87,7 +88,7 @@ class AttentionLSTM(ModelBase):
         for i, (input_dim, feature
                 ) in enumerate(zip(self.feature_dims, self.feature_input)):
             att = LSTMAttentionModel(input_dim, self.embedding_size,
-                                     self.lstm_size)
+                                     self.lstm_size, self.drop_rate)
             att_out = att.forward(feature, is_training=(self.mode == 'train'))
             att_outs.append(att_out)
         out = fluid.layers.concat(att_outs, axis=1)

fluid/PaddleCV/video/models/attention_lstm/lstm_attention.py

@@ -27,9 +27,7 @@ class LSTMAttentionModel(object):
                  drop_rate=0.5):
         self.lstm_size = lstm_size
         self.embedding_size = embedding_size
-        self.bias_attr = ParamAttr(
-            regularizer=fluid.regularizer.L2Decay(0.0),
-            initializer=fluid.initializer.NormalInitializer(scale=0.0))
+        self.drop_rate = drop_rate
 
     def forward(self, input, is_training):
         input_fc = fluid.layers.fc(
@@ -64,7 +62,7 @@ class LSTMAttentionModel(object):
             input=[lstm_forward, lstm_backward], axis=1)
 
         lstm_dropout = fluid.layers.dropout(
-            x=lstm_concat, dropout_prob=0.5, is_test=(not is_training))
+            x=lstm_concat, dropout_prob=self.drop_rate, is_test=(not is_training))
 
         lstm_weight = fluid.layers.fc(
             input=lstm_dropout,