在训练时出现TypeError: simple_img_conv_pool() got an unexpected keyword argument 'stride'

• 关键字：stride，步长

• 问题描述：在使用fluid.nets.simple_img_conv_pool()接口建立一个卷积神经网络时，当通过参数stride设置卷积操作的滑动步长，在训练的时候报错，提示stride参数不存在。

• 报错信息：

<ipython-input-7-b3ae5da446df> in main()
     10 
     11     trainer = Trainer(
---> 12         train_func=train_program, place=place, optimizer_func=optimizer_program)
     13 
     14     # Save the parameter into a directory. The Inferencer can load the parameters from it to do infer

/usr/local/lib/python3.5/dist-packages/paddle/fluid/contrib/trainer.py in __init__(self, train_func, optimizer_func, param_path, place, parallel, checkpoint_config)
    257 
    258         with framework.program_guard(self.train_program, self.startup_program):
--> 259             program_func_outs = train_func()
    260             self.train_func_outputs = program_func_outs if isinstance(
    261                 program_func_outs, list) else [program_func_outs]

<ipython-input-6-e0d473e7889c> in train_program()
      5     # predict = softmax_regression() # uncomment for Softmax
      6     # predict = multilayer_perceptron() # uncomment for MLP
----> 7     predict = convolutional_neural_network()  # uncomment for LeNet5
      8 
      9     # Calculate the cost from the prediction and label.

<ipython-input-4-0966b62f60c9> in convolutional_neural_network()
      9         pool_size=2,
     10         pool_stride=2,
---> 11         act="relu")
     12     conv_pool_1 = fluid.layers.batch_norm(conv_pool_1)
     13     # second conv pool

TypeError: simple_img_conv_pool() got an unexpected keyword argument 'stride'

• 问题复现：使用fluid.nets.simple_img_conv_pool()定义一个卷积神经网络，并使用stride参数设置卷积操作的滑动步长。最后使用这个卷积神经网络进行训练，便出现该问题。错误代码如下：

def convolutional_neural_network():
    img = fluid.layers.data(name='img', shape=[1, 28, 28], dtype='float32')
    # first conv pool
    conv_pool_1 = fluid.nets.simple_img_conv_pool(
        input=img,
        filter_size=5,
        num_filters=20,
        stride=1,
        pool_size=2,
        pool_stride=2,
        act="relu")
    conv_pool_1 = fluid.layers.batch_norm(conv_pool_1)
    # second conv pool
    conv_pool_2 = fluid.nets.simple_img_conv_pool(
        input=conv_pool_1,
        filter_size=5,
        num_filters=50,
        stride=1,
        pool_size=2,
        pool_stride=2,
        act="relu")
    prediction = fluid.layers.fc(input=conv_pool_2, size=10, act='softmax')
    return prediction

• 问题解决：错误的原因是fluid.nets.simple_img_conv_pool()接口没有stride这个参数，如果需要设置卷积操作的滑动步长，可以使用这个paddle.fluid.layers.conv2d()接口，这个接口有stride参数可以设置卷积操作的步长。

def convolutional_neural_network():
    img = fluid.layers.data(name='img', shape=[1, 28, 28], dtype='float32')
    # first conv pool
    conv_pool_1 = fluid.nets.simple_img_conv_pool(
        input=img,
        filter_size=5,
        num_filters=20,
        pool_size=2,
        pool_stride=2,
        act="relu")
    conv_pool_1 = fluid.layers.batch_norm(conv_pool_1)
    # second conv pool
    conv_pool_2 = fluid.nets.simple_img_conv_pool(
        input=conv_pool_1,
        filter_size=5,
        num_filters=50,
        pool_size=2,
        pool_stride=2,
        act="relu")
    prediction = fluid.layers.fc(input=conv_pool_2, size=10, act='softmax')
    return prediction

• 问题拓展：卷积操作的需要使用到的参数有：滑动步长(stride)、填充长度(padding)、卷积核窗口大小(filter size)、分组数(groups)、扩张系数(dilation rate)。针对卷积操作，PaddlePaddle提供了这个接口paddle.fluid.layers.conv2d。