在 Tensorflow 2.0 中使用我的自定义嵌入层时出错

Question

I've made my custom embedding layer but when I'm testing it I'm getting error. Below is my custom embedding layer.

class EndTokenLayer(Layer):

  def __init__(self, embedding_dim=128, **kwargs):
    super(EndTokenLayer, self).__init__(**kwargs)
    self.end_token_embedding = tf.Variable(initial_value=tf.random.uniform(shape=(embedding_dim,)), trainable=True)

  def call(self, inputs):
    end_token = tf.tile(tf.reshape(self.end_token_embedding, shape=(1, 1, self.end_token_embedding.shape[0])), [tf.shape(inputs)[0],1,1])
    return tf.keras.layers.concatenate([inputs, end_token], axis=1)

But when I'm testing to my train_dataset(from tensorflow slices) which is having shape of one batch x = (16,13,128) and y = (16,14)

temp = EndTokenLayer()
print(temp(inputs = train.take(1)))

Error logs: ValueError: Attempt to convert a value () with an unsupported type () to a Tensor.

Answer 1

train.take(1) will give you a new dataset with a single element, not the first element of the dataset. Maybe you want something like this:

temp = EndTokenLayer()
for elem in train.take(1):
    print(temp(inputs=elem))

Or just:

temp = EndTokenLayer()
print(temp(inputs=next(iter(train))))