问题描述
我试图弄清楚stop_gradient在Keras中的工作原理。
Python 3.7,Tensorflow 1.14.0,Keras 2.2.4
ValueError: An operation has `None` for gradient. Please make sure that all of your ops have a gradient defined (i.e. are differentiable). Common ops without gradient: K.argmax,K.round,K.eval.
这是我的代码:
input_layer = Input(shape = train_set[0].shape)
hidden_layer1 = Dense(nbr_f,activation='sigmoid')(input_layer)
lambda_layer_stopGrad = Lambda(lambda hidden_layer1: K.stop_gradient(hidden_layer1))(hidden_layer1)
hidden_layer2 = Dense((nbr_f + 1 + Y_train_cat.shape[1]),activation='sigmoid')(lambda_layer_stopGrad)
output_layer = Dense(Y_train_cat.shape[1],activation='softmax')(hidden_layer2)
model = Model(inputs = input_layer,outputs = output_layer)
model.compile(optimizer = 'SGD',loss = 'categorical_crossentropy',metrics = ['accuracy'])
model.fit(train_set,Y_train_cat,validation_data = (valid_set,Y_valid_cat),epochs = nbr_of_epochs,verbose = 0)
实际上,我希望某个层的某些神经元不参与反向传播,但是我从一个简单的任务开始:防止整个层沉积到反向传播。
解决方法
暂无找到可以解决该程序问题的有效方法,小编努力寻找整理中!
如果你已经找到好的解决方法,欢迎将解决方案带上本链接一起发送给小编。
小编邮箱:dio#foxmail.com (将#修改为@)