python - 访问由tf.keras创建的掩码(Accessing mask created by tf.keras)

Question

When using masks, for examples with tf.keras.layers.Embedding(..., mask_zero=True) , a mask is created automatically for each element which equals to 0.

(使用遮罩时，例如使用tf.keras.layers.Embedding(..., mask_zero=True) ，将自动为每个等于0的元素创建一个遮罩。)

Since once we moved through the layer, the mask is already computed, is there any way to access it?

(由于一旦我们移动了图层，就已经计算出了蒙版，是否有任何方法可以访问它？)

(without calling compute_mask again)

(（无需再次调用compute_mask ）)

For example, consider the following code

(例如，考虑以下代码)

inputs = np.random.randint(0,999, (10,10))
inputs[inputs < 100] = 0  # Create some 0 values

embedding_layer = tf.keras.layers.Embedding(1000, 64, input_length=10, mask_zero=True)
result = embedding_layer(inputs)  # Mask is already calculated here

### How to access the mask here?

There is a private variable result._keras_mask , but I assume since it is marked private I better not interact with it.

(有一个私有变量result._keras_mask ，但是我假设由于将其标记为私有，所以最好不要与之交互。)

As I don't want to call embedding_layer.compute_mask (since it was already called) I'm wondering how should I access the mask?

(因为我不想调用embedding_layer.compute_mask （因为它已经被调用），所以我想知道应该如何访问该掩码？)

I know it is passed automatically to sequential layers, but I want to access it anyway

(我知道它会自动传递到顺序图层，但无论如何我都想访问它)

  ask by bluesummers translate from so

Answer 1

等待大神答复