Keras层的稀疏输出

我有一个预测某些系数的网络。以下是我的模型的代码：

ink = Flatten()(in1)
ink = Dense(units=units,activation = activation)(ink)
ink = Dense(units=units,activation = activation)(ink) 
ink = Dense(units=units,activation = activation)(ink) 
coeff = Dense(units=nout,activation = None,name='coeffs')(ink)

in2 = Input(shape=(x_train.shape[1],1),name = 'in_2')
in2 = Flatten()(in2)
in2 = Dense(units=units,activation = activation)(in2)
in2 = Dense(units=units,activation = activation)(in2) 
in2 = Dense(units=units,activation = activation)(in2)
in2 = Dense(units=nout,activation = None)(in2) 
prediction  = keras.layers.Dot(axis =1 )([in2,coeff])}

所以我希望层系数的输出是稀疏的。我已经将带有L1的activity_正则化器应用于该层，但它似乎不起作用。 提前谢谢