fix if logic for dygraph, test=develop (#4447)

dygraph/bert/model/transformer_encoder.py

@@ -37,7 +37,7 @@ class PrePostProcessLayer(Layer):
 
         for cmd in self.process_cmd:
             if cmd == "a":  # add residual connection
-                self.functors.append(lambda x, y: x + y if y else x)
+                self.functors.append(lambda x, y: x + y if y is not None else x)
                 self.exec_order += "a"
             elif cmd == "n":  # add layer normalization
                 self.functors.append(
@@ -215,7 +215,7 @@ class MultiHeadAttentionLayer(Layer):
             y=transpose_k,
             transpose_y=True)
         #alpha=self._d_model**-0.5)
-        if attn_bias:
+        if attn_bias is not None:
             product += attn_bias
         weights = fluid.layers.softmax(product)
         if self._dropout_rate:

dygraph/lac/sequence_labeling.py

@@ -164,7 +164,7 @@ class Linear_chain_crf(fluid.dygraph.Layer):
             "Transition": self._transition,
             "Label": [label]
         }
-        if length:
+        if length is not None:
             this_inputs['Length'] = [length]
         self._helper.append_op(
                         type='linear_chain_crf',
@@ -212,7 +212,7 @@ class Crf_decoding(fluid.dygraph.Layer):
         viterbi_path = self._helper.create_variable_for_type_inference(
                         dtype=self._dtype)
         this_inputs = {"Emission": [input], "Transition": self._transition, "Label": label}
-        if length:
+        if length is not None:
             this_inputs['Length'] = [length]
         self._helper.append_op(
                         type='crf_decoding',
@@ -245,7 +245,7 @@ class Chunk_eval(fluid.dygraph.Layer):
         num_correct_chunks = self._helper.create_variable_for_type_inference(dtype="int64")
 
         this_input = {"Inference": [input], "Label": [label]}
-        if seq_length:
+        if seq_length is not None:
             this_input["SeqLength"] = [seq_length]
 
         self._helper.append_op(

dygraph/sentiment/nets.py

@@ -156,7 +156,7 @@ class BOW(fluid.dygraph.Layer):
         fc_1 = self._fc1(bow_1)
         fc_2 = self._fc2(fc_1)
         prediction = self._fc_prediction(fc_2)
-        if label:
+        if label is not None:
             cost = fluid.layers.cross_entropy(input=prediction, label=label)
             avg_cost = fluid.layers.mean(x=cost)
             acc = fluid.layers.accuracy(input=prediction, label=label)
@@ -258,4 +258,4 @@ class BiGRU(fluid.dygraph.Layer):
             acc = fluid.layers.accuracy(input=prediction, label=label)
             return avg_cost, prediction, acc
         else:
-            return prediction
\ No newline at end of file
+            return prediction

dygraph/seq2seq/attention_model.py

@@ -169,7 +169,7 @@ class AttentionModel(fluid.dygraph.Layer):
         memory = self.attn_fc(enc_output)
         attn = fluid.layers.matmul(query, memory, transpose_y=True)
 
-        if mask:
+        if mask is not None:
             attn = fluid.layers.transpose(attn, [1, 0, 2])
             attn = fluid.layers.elementwise_add(attn, mask * 1000000000, -1)
             attn = fluid.layers.transpose(attn, [1, 0, 2])

dygraph/transformer/model.py

@@ -78,7 +78,7 @@ class PrePostProcessLayer(Layer):
         self.functors = []
         for cmd in self.process_cmd:
             if cmd == "a":  # add residual connection
-                self.functors.append(lambda x, y: x + y if y else x)
+                self.functors.append(lambda x, y: x + y if y is not None else x)
             elif cmd == "n":  # add layer normalization
                 self.functors.append(
                     self.add_sublayer(
@@ -156,7 +156,7 @@ class MultiHeadAttention(Layer):
                                 y=k,
                                 transpose_y=True,
                                 alpha=self.d_model**-0.5)
-        if attn_bias:
+        if attn_bias is not None:
             product += attn_bias
         weights = layers.softmax(product)
         if self.dropout_rate:
@@ -936,4 +936,4 @@ class Transformer(Layer):
             layers.gather_tree(predict_ids, parent_ids), [1, 2, 0])
         finished_scores = topk_scores
 
-        return finished_seq, finished_scores
\ No newline at end of file
+        return finished_seq, finished_scores