x1 = Dense(300, activation='relu', kernel_initializer=initializers.glorot_uniform(seed=seed))(x0)
p1 = Dense(nclass, activation='sigmoid')(x1)
x1_st = Lambda(lambda x: K.stop_gradient(x))(x1)
# x1_st=MyLayer()(x1_st)
p1_st = Lambda(lambda x: K.stop_gradient(x))(p1)
x2 = concatenate([x0, x1_st, p1_st])
x2 = Dense(300, activation='relu', kernel_initializer=initializers.glorot_uniform(seed=seed))(x2)
p2 = Dense(nclass, activation='sigmoid', kernel_initializer=initializers.glorot_uniform(seed=seed))(x2)
I wonder how does stop_gradient work here. What will happen if I just remove the two Lambda layer?