keras_contrib使用问题

1、keras_contrib安装

https://github.com/keras-team/keras-contrib

下载后将keras_contrib文件夹放到python的\Lib\site-packages里面

2、报错The added layer must be an instance of class Layer. Found: <keras_contrib.la

要统一导入方式，tensorflow.keras、keras引用混合

3、统一用keras导入

报错module ‘tensorflow.python.framework.ops’ has no attribute ‘_TensorLike’

原因：TensorFlow和keras版本不统一报错，原tf版本2.4，keras版本2.3.1，将tf降低到2.1.0

4、报错Tensors in list passed to ‘values’ of ‘ConcatV2’ Op have types [bool, float3

参考https://blog.csdn.net/u011740601/article/details/103800575

module ‘tensorflow’ has no attribute ‘to_float’

加上tf.to_float = lambda x: tf.cast(x, tf.float32) 参考https://ask.csdn.net/questions/2962902

修改keras_contrib的keras_contrib\layers\cfr.py，

import tensorflow as tf
tf.to_float = lambda x: tf.cast(x, tf.float32)
if mask is not None:
# mask2 = K.cast(K.concatenate([mask, K.zeros_like(mask[:, :1])], axis=1),
# K.floatx())#改为下面的
mask2 = K.cast(K.concatenate([tf.to_float(mask), K.zeros_like(mask[:, :1])], axis=1),
K.floatx())
constants.append(mask2)