keras 升级_Debug 路漫漫-08：Keras 版本升级函数变换导致的问题

在使用 CNN的时候，报错： TypeError: ('Keyword argument not understood:', 'padding')

将“padding”改为“border_mode”，即可：

原因：padding 是Keras 2.X的语法，而我的PC安装的是 Keras 1.X版本。

二者的API 有一些地方是有变化的。

如下：(从 1.X 到 2.X )

========【Models】

1、Constructor arguments for Model have been renamed：

input -> inputs

output -> outputs

2、The Sequential model not longer supports the set_input method.

3、For any model saved with Keras 2.0 or higher, weights trained with backend X will be converted to work with backend Y without any manual conversion step.

========【Layers】

1、Dense layer

Changed interface：

output_dim -> units

init -> kernel_initializer

added bias_initializer argument

W_regularizer -> kernel_regularizer

b_regularizer -> bias_regularizer

b_constraint -> bias_constraint

bias -> use_bias

2、Embedding

Convolutional layers ：

Interface changes common to all convolutional layers：

nb_filter -> filters

float kernel dimension arguments become a single tuple argument, kernel size. E.g. a legacy call Conv2D(10, 3, 3) becomes Conv2D(10, (3, 3))

kernel_size can be set to an integer instead of a tuple, e.g. Conv2D(10, 3) is equivalent toConv2D(10, (3, 3)).

subsample -> strides. Can also be set to an integer.

border_mode -> padding

init -> kernel_initializer

added bias_initializer argument

W_regularizer -> kernel_regularizer

b_regularizer -> bias_regularizer

b_constraint -> bias_constraint

bias -> use_bias

dim_ordering -> data_format

In the SeparableConv2D layers, init is split into depthwise_initializer andpointwise_initializer.

Added dilation_rate argument in Conv2D and Conv1D.

1D convolution kernels are now saved as a 3D tensor (instead of 4D as before).

2D and 3D convolution kernels are now saved in format spatial_dims + (input_depth, depth)), even with data_format="channels_first".

3、Pooling1D

pool_length -> pool_size

stride -> strides

border_mode -> padding

4、Pooling2D，3D

border_mode -> padding

dim_ordering -> data_format

【Reference】