PaddlePaddle在执行词向量预测的是出现预测数据不能完全转换为Python ndarray的错误

关键字：预测，numpy
问题描述：通过使用fluid.create_lod_tensor创建一个预测数据，然后使用预测器对数据进行预测，在执行预测的时候，报预测的张量数据中有一些包含LoD信息，它们不能完全转换为Python ndarray的错误。
报错信息：

<ipython-input-9-196da4d402b0> in infer(use_cuda, inference_program, params_dirname)
     27             'secondw': second_word,
     28             'thirdw': third_word,
---> 29             'fourthw': fourth_word
     30         })
     31 

······
/usr/local/lib/python3.5/dist-packages/paddle/fluid/contrib/inferencer.py in infer(self, inputs, return_numpy)
    102             results = self.exe.run(feed=inputs,
    103                                    fetch_list=[self.predict_var.name],
--> 104                                    return_numpy=return_numpy)
    105 
    106         return results

/usr/local/lib/python3.5/dist-packages/paddle/fluid/executor.py in run(self, program, feed, fetch_list, feed_var_name, fetch_var_name, scope, return_numpy, use_program_cache)
    471         outs = self._fetch_data(fetch_list, fetch_var_name, scope)
    472         if return_numpy:
--> 473             outs = as_numpy(outs)
    474         return outs

/usr/local/lib/python3.5/dist-packages/paddle/fluid/executor.py in as_numpy(tensor)
     83         return [as_numpy(t) for t in tensor]
     84     if isinstance(tensor, list):
---> 85         return [as_numpy(t) for t in tensor]
     86     assert isinstance(tensor, core.LoDTensor)
     87     lod = tensor.lod()

/usr/local/lib/python3.5/dist-packages/paddle/fluid/executor.py in <listcomp>(.0)
     83         return [as_numpy(t) for t in tensor]
     84     if isinstance(tensor, list):
---> 85         return [as_numpy(t) for t in tensor]
     86     assert isinstance(tensor, core.LoDTensor)
     87     lod = tensor.lod()

/usr/local/lib/python3.5/dist-packages/paddle/fluid/executor.py in as_numpy(tensor)
     90             They can not be completely cast to Python ndarray. \
     91             Please set the parameter 'return_numpy' as 'False' to \
---> 92             return LoDTensor itself directly.")
     93     return np.array(tensor)
     94 

RuntimeError: Some of your fetched tensors hold LoD information.             They can not be completely cast to Python ndarray.             Please set the parameter 'return_numpy' as 'False' to             return LoDTensor itself directly.

问题复现：使用定义的网络和训练过程中保存的模型参数创建一个预测器，然后再使用fluid.create_lod_tensor创建4个预测数据，最后使用预测器对4个预测数据进行预测，没有设置参数return_numpy，在执行的时候就报错。错误代码如下：

first_word = fluid.create_lod_tensor(data1, lod, place)
second_word = fluid.create_lod_tensor(data2, lod, place)
third_word = fluid.create_lod_tensor(data3, lod, place)
fourth_word = fluid.create_lod_tensor(data4, lod, place)

result = inferencer.infer(
    {
        'firstw': first_word,
        'secondw': second_word,
        'thirdw': third_word,
        'fourthw': fourth_word
    })

解决问题：在对词向量进行预测时，返回的结果并不是一个numpy值，而接口paddle.fluid.contrib.inferencer.infer默认的返回值的是一个numpy类型的，所以就报错。需要设置参数return_numpy的值为False。正确代码如下：

first_word = fluid.create_lod_tensor(data1, lod, place)
second_word = fluid.create_lod_tensor(data2, lod, place)
third_word = fluid.create_lod_tensor(data3, lod, place)
fourth_word = fluid.create_lod_tensor(data4, lod, place)

result = inferencer.infer(
    {
        'firstw': first_word,
        'secondw': second_word,
        'thirdw': third_word,
        'fourthw': fourth_word
    },
    return_numpy=False)

问题拓展：接口paddle.fluid.contrib.inferencer.infer是属于高层接口，与低层接口不同的是，高层接口不需要用户操作执行器。高层接口使用更简单，虽然没有低层接口灵活，但是使用方便，适合初学者使用。

PaddlePaddle在执行词向量预测的是出现预测数据不能完全转换为Python ndarray的错误

猜你喜欢