Created by: CDDB

work的网络结构：

from paddle.trainer_config_helpers import *

is_predict = get_config_arg("is_predict", bool, False)
a_dim = 60
b_dim = 80
num_classes = 2

####################Data Configuration ##################
if not is_predict:
  data_dir='dna_data/'
  args = {'a_dim':a_dim, 'b_dim':b_dim, 'num_classes': num_classes}
  define_py_data_sources2(train_list=data_dir+"train.list",
                          test_list=data_dir+'test.list',
                          module='ei_fea_provider',
                          obj='processData',
                          args=args)

######################Algorithm Configuration #############
settings(
    batch_size = 500,
    learning_method = AdaDeltaOptimizer()
)
#######################Network Configuration #############
a_input = data_layer(name='a_fea', size=a_dim)
b_input = data_layer(name='b_fea', size=b_dim)
label = data_layer(name="label", size=num_classes)

a_hidden = fc_layer(input=a_input, size=256, act=ReluActivation())
b_hidden = fc_layer(input=b_input, size=256, act=ReluActivation())

**prediction = fc_layer(input=[a_hidden, b_hidden], size=num_classes, act=SoftmaxActivation())**
outputs(classification_cost(input=prediction,
                            label=label))

其中，调换prediction layer的input的顺序，就会出现check失败的情况。错误的一行配置 prediction = fc_layer(input=[b_hidden, a_hidden], size=num_classes, act=SoftmaxActivation())

错误log：

[INFO 2016-11-01 17:07:42,249 networks.py:1129] The output order is [__cost_0__]
I1101 17:07:42.251242  3502 Trainer.cpp:169] trainer mode: Normal
I1101 17:07:42.252277  3502 PyDataProvider2.cpp:219] loading dataprovider ei_fea_provider::processData
I1101 17:07:42.252609  3502 PyDataProvider2.cpp:219] loading dataprovider ei_fea_provider::processData
I1101 17:07:42.252691  3502 GradientMachine.cpp:134] Initing parameters..
I1101 17:07:42.255108  3502 GradientMachine.cpp:141] Init parameters done.
F1101 17:07:47.468036  3502 Matrix.cpp:2015] Check failed: a_col == b_row (80 vs. 60)
*** Check failure stack trace: ***
    @     0x7f918ba5852d  google::LogMessage::Fail()
    @     0x7f918ba5cc90  google::LogMessage::SendToLog()
    @     0x7f918ba59057  google::LogMessage::Flush()
    @     0x7f918ba592b6  google::LogMessageFatal::~LogMessageFatal()
    @           0x7eb2b9  paddle::CpuMatrix::mul()
    @           0x7f324b  paddle::CpuMatrix::mul()
    @           0x64b811  paddle::FullyConnectedLayer::forward()
    @           0x6c7bc6  paddle::NeuralNetwork::forward()
    @           0x57eebc  paddle::GradientMachine::forwardBackward()
    @           0x70f1e5  paddle::TrainerInternal::forwardBackwardBatch()
    @           0x710fe5  paddle::TrainerInternal::trainOneBatch()
    @           0x70a66a  paddle::Trainer::trainOnePass()
    @           0x70b553  paddle::Trainer::train()
    @           0x537794  main
    @       0x318ae1ecdd  (unknown)
    @           0x53faf5  (unknown)

paddle版本
PaddlePaddle 0.8.0b0, compiled with
    with_avx: ON
    with_gpu: OFF
    with_double: OFF
    with_python: ON
    with_rdma: OFF
    with_glog: ON
    with_gflags: ON
    with_metric_learning: OFF
    with_timer: OFF
    with_predict_sdk: OFF