Question

我是DL的新手，并将千层面CNN重新编码为Keras（TF）

该层的一部分是Conv1D，然后是Maxout，其特征为maxout

Maxout函数已从keras2.0中删除

我参考了stack overflow和git-hub 编写自定义lambda函数

def Maxout(x, num_unit=None):
    input_shape = x.get_shape().as_list()
    ch = input_shape[-1]
    num_unit = int(ch / 2)
    assert ch is not None and ch % num_unit == 0
    x = K.backend.reshape(x, (-1, ch // int(num_unit) , int(num_unit)))
    x = K.backend.max(x, axis=1,keepdims=True)


input_tensor = Input(shape=(128,32),name = 'input')
conv4= (Conv1D(64, kernel_size=5, strides=1,
                 padding = 'same',
                 name = 'conv4',
                 input_shape=(128,32)))(maxpool1)
output = Lambda(Maxout,name='maxout')(conv4)

model = Model(inputs=input_tensor, outputs=output)
print(model.summary())

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input (InputLayer)           (None, 128, 32)           0         
_________________________________________________________________
conv4 (Conv1D)               (None, 128, 64)           20544     
_________________________________________________________________
maxout (Lambda)              (None, 1, 32)             0         
=================================================================

我期望从输入：（None，128,64）到输出：（None，128,32）的Maxout层

如何在128,32中获得输出的形状？

如何为Conv1D输出实施Maxout激活

0 个答案: