Keras - 獲得訓練層的權重

Question

我試圖獲得訓練網絡中圖層的值。我可以得到層作爲TensorFlow張量，但我無法訪問陣列狀的價值觀：

from keras.models import load_model 

model = load_model('./model.h5') 
layer_dict = dict([(layer.name, layer) for layer in model.layers]) 

layer_name = 'block5_sepconv1_act' 
filter_index = 0 

layer_output = layer_dict['model_1'][layer_name].output 
# <tf.Tensor 'block5_sepconv1_act/Relu:0' shape=(?, 16, 16, 728) dtype=float32> 
layer_filter = layer_output[:, :, :, filter_index] 
# <tf.Tensor 'strided_slice_11:0' shape=(?, 16, 16) dtype=float32> 
# how do I get the 16x16 values ?? 

Answer 1

如果使用tensorflow後端，您可以使用當前會話sess和餵養正確的輸入

import keras.backend as K 

input_value = np.zeros(size=(batch_size, input_dim)) 
sess = K.get_session() 
output = sess.run(layer_output, feed_dict={model.input: input_value}) 

評價張量的值。如果你只是想取回權，你可以評價的權重一層使用：

weights = [w.eval(K.get_session) for w in layer_dict['model_1'][layer_name].weights] 

Answer 2

.get_weights()會返回一個特定層或模型的權重作爲numpy的陣列

layer_dict[layer_name].get_weights() 

如果您想要圖層的輸出，請檢查here問題上的答案。