嵌入层输出形状为2D

Question

我在嵌入层的输出形状上遇到了一些问题，根据keras文档，嵌入层应具有3D张量的输出形状，但是我的嵌入层仅输出2D张量。

class MyModel(Model):
  def __init__(self, vocab_size, embedding_matrix, max_length):
      super(MyModel, self).__init__()
      self.embedding_l1 = tf.keras.layers.Embedding(input_dim=vocab_size,
                                                    output_dim=max_length, 
                                                    input_length=max_length,
                                                    weights=[embedding_matrix], 
                                                    trainable=False)
      self.bidirectional_l1 = Bidirectional(
                             tf.compat.v1.keras.layers.CuDNNLSTM(32, 
                                                                 return_sequences=False))
      self.dense_l1 = Dense(units=256, activation='relu')
      self.dropout_l1 = Dropout(rate=2e-5)
      self.dense_l2 = Dense(units=1, activation='sigmoid')

  def call(self, x):
      embedding_out = self.embedding_l1(x)
      print("SHAPE:",embedding_out.shape)
      bid_out1 = self.bidirectional_l1(self.reshape_l1(embedding_out))
      dense_out1 = self.dense_l1(bid_out1)
      drop_out1 = self.dropout_l2(dense_out1)
      dense_out2 = self.dense_l2(drop_out2)
      return dense_out2

将嵌入层的形状输出为2D（300,300）张量。这会导致双向lstm错误：

ValueError: Input 0 of layer bidirectional is incompatible with the layer: expected ndim=3, found ndim=2. Full shape received: [300, 300]

Answer 1

找出它是嵌入层上的输入。我没有提供批处理大小，因此嵌入的输入看起来是[batch_size，300]，而不是[300]。