Created by: qingqing01

存在的问题：

• 网络拓扑解析了2次：paddle.parameters.create(cost) 和 trainer.train中。
• 目前CPU和GPU训练classification_error_evaluator降的都不太正常~

python api_train_v2.py
I0227 23:29:45.882282  6005 Util.cpp:160] commandline:  --use_gpu=True --trainer_count=1
[INFO 2017-02-27 23:29:50,170 networks.py:1472] The input order is [pixel, label]
[INFO 2017-02-27 23:29:50,170 networks.py:1478] The output order is [__classification_cost_0__]
[INFO 2017-02-27 23:29:50,176 networks.py:1472] The input order is [pixel, label]
[INFO 2017-02-27 23:29:50,176 networks.py:1478] The output order is [__classification_cost_0__]
I0227 23:29:50.209223  6005 GradientMachine.cpp:86] Initing parameters..
I0227 23:29:50.227862  6005 GradientMachine.cpp:93] Init parameters done.
Pass 0, Batch 0, Cost 2.685198, {'classification_error_evaluator': 0.8984375}
Pass 0, Batch 100, Cost 0.542156, {'classification_error_evaluator': 0.1875}

还会继续调试，同时写srl的demo~