Keras切换backend ： theano --> tensorflow

1、切换backend
修改~/.keras/keras.json 文件中的 theano 字段为tensorflow即可
官方文档： https://keras.io/backend/


2、theano和tensorflow卷积核互相转换
切换backend后，模型运算会出错，原因在于tensorflow中的卷积实际上时相关，二theano中的卷积是真正的卷积！！！！所以，切换backend时，需要对卷积核进行翻转

参见： https://github.com/fchollet/keras/wiki/Converting-convolution-kernels-from-Theano-to-TensorFlow-and-vice-versa

通用转换代码如下（theano和tensorflow互转）：

from keras import backend as K
from keras.utils.np_utils import convert_kernel

model = model_from_json(open(os.path.join('.', 'model.json')).read())
model.load_weights(os.path.join('.',  'model_weights.h5'))



for layer in model.layers:
   if layer.__class__.__name__ in ['Convolution1D', 'Convolution2D','Convolution3D', 'AtrousConvolution2D']:
      original_w = K.get_value(layer.W)
      converted_w = convert_kernel(original_w)
      K.set_value(layer.W, converted_w)


print('running')   
K.get_session().run(ops)
print('saving')    
model.save_weights('model_weights_anotherBackend.h5')

tensoflow专用转换代码如下：

from keras import backend as K
from keras.utils.np_utils import convert_kernel
import tensorflow as tf

model = model_from_json(open(os.path.join('.', 'model.json')).read())
model.load_weights(os.path.join('.',  'model_weights.h5'))
ops = []
for layer in model.layers:
    if layer.__class__.__name__ in ['Convolution1D', 'Convolution2D', 'Convolution3D', 'AtrousConvolution2D']:
        original_w = K.get_value(layer.W)
        print(layer.W.name)
        print('\t',end='')
        print(layer.W.get_shape().to_list())
        converted_w = convert_kernel(original_w)
        ops.append(tf.assign(layer.W, converted_w).op)

print('running')   
K.get_session().run(ops)
print('saving')    
model.save_weights('model_weights_tensorflow.h5')