Created by: hwhwhwaaaa

在实现attention网络时报错，代码如下：

def get_p_attention(p_emb_layer, lstm_layer, hidden_dim):
    p_fc = fluid.layers.fc(input=p_emb_layer, size=hidden_dim, act='tanh')
    plstm_0, _ = fluid.layers.dynamic_lstm(input=p_fc, \
        size=hidden_dim,
        candidate_activation='relu',
        gate_activation='sigmoid',
        cell_activation='sigmoid',
        is_reverse=True)
    p_lstm_layer = fluid.layers.sequence_last_step(input=plstm_0)
    p_expand = fluid.layers.sequence_expand(x=p_lstm_layer, y=lstm_layer)
    combined_input = fluid.layers.elementwise_mul(x=lstm_layer,
                                        y=p_expand)
    attention_weight = fluid.layers.fc(input=combined_input,
                                        size=1,
                                        act='tanh',
                                        bias_attr=False)
    #得到归一化权重
    normed_attention_weight = fluid.layers.sequence_softmax(input=attention_weight)
    assist_info = {'unnormalized_p_attention_weight': attention_weight}
    return normed_attention_weight, assist_info

def att_model():
    word_emb_fixed = True if conf_dict['word_emb_fixed'] == "True" else False
    emb_distributed = not conf_dict['is_local']
    conf_dict['is_sparse'] = bool(conf_dict['is_sparse'])
    char_param = fluid.ParamAttr(name=conf_dict['emb_name'],
                                 trainable=(not char_emb_fixed))
    
    char_embedding = fluid.layers.embedding(
        input=char,
        size=[data_reader.get_dict_size('charemb_dict'),
            conf_dict['char_dim']],
        dtype='float32',
        is_distributed=emb_distributed,
        is_sparse=emb_distributed,
        param_attr=char_param)

    p_embedding = fluid.layers.embedding(
        input=p_word,
        size=[data_reader.get_dict_size('charemb_dict'),
            conf_dict['char_dim']],
        dtype='float32',
        is_distributed=emb_distributed,
        is_sparse=emb_distributed,
        param_attr=char_param)

    emb_layers = [char_embedding, p_embedding]
    # input hidden
    char_fc = fluid.layers.fc(input=char_embedding, size=hidden_dim, act='tanh')
    lstm_layer, _ = fluid.layers.dynamic_lstm(
        input=char_fc,
        size=hidden_dim,
        candidate_activation='relu',
        gate_activation='sigmoid',
        cell_activation='sigmoid',
        is_reverse=0)
    position_weight_layer, position_attention_assist_info = \
        get_p_attention(p_embedding, lstm_layer, hidden_dim)
    p_scaled_lstm_layer = fluid.layers.elementwise_mul(x=lstm_layer,
                                    y=position_weight_layer)
    p_lstm_layer = fluid.layers.sequence_pool(input=p_scaled_lstm_layer,
                                            pool_type='sum')
  
    lstm_layers = [p_lstm_layer, lstm_layer]
    hidden_0_layers = [
        fluid.layers.fc(input=l_layer, size=hidden_dim, act='tanh')
            for l_layer in lstm_layers
        ]
    hidden_0 = fluid.layers.sums(input=hidden_0_layers)
    lstm_0 = fluid.layers.dynamic_lstm(
        input=hidden_0/4,
        size=hidden_dim,
        candidate_activation='relu',
        gate_activation='sigmoid',
        cell_activation='sigmoid')

    # stack L-LSTM and R-LSTM with direct edges
    input_tmp = [hidden_0, lstm_0]

    for i in range(1, depth):
        mix_hidden = fluid.layers.sums(input=[
            fluid.layers.fc(input=input_tmp[0], size=hidden_dim, act='tanh'),
            fluid.layers.fc(input=input_tmp[1], size=hidden_dim, act='tanh')
        ])

        lstm = fluid.layers.dynamic_lstm(
            input=mix_hidden,
            size=hidden_dim,
            candidate_activation='relu',
            gate_activation='sigmoid',
            cell_activation='sigmoid',
            is_reverse=((i % 2) == 1))

        input_tmp = [mix_hidden, lstm]

    # output
    feature_out = fluid.layers.sums(input=[
        fluid.layers.fc(input=input_tmp[0], size=label_dict_len, act='tanh'),
        fluid.layers.fc(input=input_tmp[1], size=label_dict_len, act='tanh')
    ])

    return feature_out
`

错误信息：
`
   cost = exe.run(main_program, feed=feeder.feed(data), fetch_list=[avg_cost])
  File "./tools/paddle_fluid/paddle_release_home/python/lib/python2.7/site-packages/paddle/fluid/executor.py", line 472, in run
    self.executor.run(program.desc, scope, 0, True, True)
paddle.fluid.core.EnforceNotMet: Enforce failed. Expected in_dim == x_dim, but received in_dim:10, 512 != x_dim:548, 512.
Input tensors must have same shape at [/paddle/paddle/fluid/operators/sum_op.cc:59]
PaddlePaddle Call Stacks:
0       0x7f78eff19ed2p paddle::platform::EnforceNotMet::EnforceNotMet(std::__exception_ptr::exception_ptr, char const*, int) + 482
1       0x7f78f044c7a5p paddle::operators::SumOp::InferShape(paddle::framework::InferShapeContext*) const + 1237
2