Fluid中使用Inferencer出现错误

原创于 2019-03-08 11:24:38 发布 · 1.2k 阅读

0 ·

CC 4.0 BY-SA版权

本文解决了一个在使用PaddlePaddle进行预测时遇到的数据类型不匹配问题，详细介绍了问题出现的原因及修改训练和预测阶段数据类型的解决方法。

部署运行你感兴趣的模型镜像

问题描述：使用Fluid版本的PaddlePaddle编写一个简单的结构时，训练没有问题，但在进行Inferencer预测操作时，报DataType of Paddle Op mul must be the same错误，我检查了自己的数据预测操作，确认了数据类型与数据结构都与Inferencer预测网络中输入层定义的数据类型与结构一致，但依旧报错
报错输出：

Traceback (most recent call last):
  File "/Users/jizhi/Desktop/Paddle/Paddlecode/code1.py", line 119, in <module>
    results = inferencer.infer({'mm': test_x})
  File "/Users/jizhi/anaconda3/envs/paddle/lib/python3.5/site-packages/paddle/fluid/contrib/inferencer.py", line 104, in infer
    return_numpy=return_numpy)
  File "/Users/jizhi/anaconda3/envs/paddle/lib/python3.5/site-packages/paddle/fluid/executor.py", line 470, in run
    self.executor.run(program.desc, scope, 0, True, True)
paddle.fluid.core.EnforceNotMet: DataType of Paddle Op mul must be the same. Get mm(5) != fc_0.w_0(6) at [/Users/paddle/minqiyang/Paddle/paddle/fluid/framework/operator.cc:847]
PaddlePaddle Call Stacks: 
0          0x10e2eaa68p paddle::platform::EnforceNotMet::EnforceNotMet(std::exception_ptr, char const*, int) + 760
1          0x10f114a10p paddle::framework::OperatorWithKernel::IndicateDataType(paddle::framework::ExecutionContext const&) const + 864
2          0x10f114aacp paddle::framework::OperatorWithKernel::GetExpectedKernelType(paddle::framework::ExecutionContext const&) const + 44
3          0x10f113099p paddle::framework::OperatorWithKernel::RunImpl(paddle::framework::Scope const&, boost::variant<paddle::platform::CUDAPlace, paddle::platform::CPUPlace, paddle::platform::CUDAPinnedPlace, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_> const&) const + 265
4          0x10f10f141p paddle::framework::OperatorBase::Run(paddle::framework::Scope const&, boost::variant<paddle::platform::CUDAPlace, paddle::platform::CPUPlace, paddle::platform::CUDAPinnedPlace, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_> const&) + 577
5          0x10e3b83a6p paddle::framework::Executor::RunPreparedContext(paddle::framework::ExecutorPrepareContext*, paddle::framework::Scope*, bool, bool, bool) + 390
6          0x10e3b7dd3p paddle::framework::Executor::Run(paddle::framework::ProgramDesc const&, paddle::framework::Scope*, int, bool, bool) + 163
7          0x10e31e837p void pybind11::cpp_function::initialize<paddle::pybind::pybind11_init()::$_64, void, paddle::framework::Executor&, paddle::framework::ProgramDesc const&, paddle::framework::Scope*, int, bool, bool, pybind11::name, pybind11::is_method, pybind11::sibling>(paddle::pybind::pybind11_init()::$_64&&, void (*)(paddle::framework::Executor&, paddle::framework::ProgramDesc const&, paddle::framework::Scope*, int, bool, bool), pybind11::name const&, pybind11::is_method const&, pybind11::sibling const&)::'lambda'(pybind11::detail::function_call&)::__invoke(pybind11::detail::function_call&) + 135
8          0x10e2f53aap pybind11::cpp_function::dispatcher(_object*, _object*, _object*) + 5786
9          0x10141659fp PyCFunction_Call + 127
10         0x1014e17e7p PyEval_EvalFrameEx + 33207
11         0x1014d7fafp _PyEval_EvalCodeWithName + 335
12         0x1014de2a7p PyEval_EvalFrameEx + 19575
13         0x1014d7fafp _PyEval_EvalCodeWithName + 335
14         0x1014de2a7p PyEval_EvalFrameEx + 19575
15         0x1014d7fafp _PyEval_EvalCodeWithName + 335
16         0x10152a758p PyRun_FileExFlags + 248
17         0x101529eeep PyRun_SimpleFileExFlags + 382
18         0x10154ed86p Py_Main + 3622
19         0x101390861p main + 497
20      0x7fff5dffe015p start + 1
21                 0x2p

问题复现：

def train_program():
    y = fluid.layers.data(name='y', shape=[1], dtype='float64')
    x = fluid.layers.data(name='x', shape=[13], dtype='float64')

    y_predict = fluid.layers.fc(input=x, size=1, act=None)
    # 平均损失
    loss = fluid.layers.square_error_cost(input=y_predict, label=y)
    avg_loss = fluid.layers.mean(loss)
    return avg_loss

trainer = Trainer(
    train_func=train_program,
    place=place,
    optimizer_func=optimizer_program
)

def inference_program():
    mm = fluid.layers.data(name='mm', shape=[13], dtype='float32')
    y_predict = fluid.layers.fc(input=mm, size=1, act=None)
    return y_predict

inferencer = Inferencer(
    infer_func = inference_program, param_path = params_dirname, place=place
)

batch_size = 10
test_reader = paddle.batch(paddle.dataset.uci_housing.test(), batch_size=batch_size)
test_data = next(test_reader()) 
test_x = numpy.array([data[0] for data in test_data]).astype('float32')
test_y = numpy.array([data[1] for data in test_data]).astype('float32')

问题分析：
但从报错输出DataType of Paddle Op mul must be the same来看，就是类型输出问题，通过问题描述中的内容来看，预测网络的结构应该是没有问题的，那么就是报数据类型问题，那么很大的可能就是训练模型的数据类型与输入的数据类型不匹配，因为预测网络要读入训练网络训练后的模型文件，所有运行网络的类型对预测时的数据类型要求也要一致。
解决方法：

将训练时，使用的数据类型与预测时输入的数据类型也对应上，就可以解决该报错，让程序正常运行。

def train_program():
    y = fluid.layers.data(name='y', shape=[1], dtype='float32')
    x = fluid.layers.data(name='x', shape=[13], dtype='float32')

    y_predict = fluid.layers.fc(input=x, size=1, act=None)
    # 平均损失
    loss = fluid.layers.square_error_cost(input=y_predict, label=y)
    avg_loss = fluid.layers.mean(loss)
    return avg_loss

trainer = Trainer(
    train_func=train_program,
    place=place,
    optimizer_func=optimizer_program
)

def inference_program():
    mm = fluid.layers.data(name='mm', shape=[13], dtype='float32')
    y_predict = fluid.layers.fc(input=mm, size=1, act=None)
    return y_predict

inferencer = Inferencer(
    infer_func = inference_program, param_path = params_dirname, place=place
)

batch_size = 10
test_reader = paddle.batch(paddle.dataset.uci_housing.test(), batch_size=batch_size)
test_data = next(test_reader()) 
test_x = numpy.array([data[0] for data in test_data]).astype('float32')
test_y = numpy.array([data[1] for data in test_data]).astype('float32')