Tensorflow1 DropoutWrapper 到 Tensorflow2

Question

如何在tensorflow2中重写这个？尝试使用

tf_upgrade_v2

但失败了。也尝试过使用

tf.keras.layers.Dropout

，但将

MultiRNNCell

转换为张量

时遇到一些问题

rnn_cells = tf.compat.v1.nn.rnn_cell.MultiRNNCell(
            [lstm_cell(size_layer) for _ in range(num_layers)],
            state_is_tuple = False,
        )
drop = tf.contrib.rnn.DropoutWrapper(rnn_cells, output_keep_prob = forget_bias)

Answer 1

这是 TensorFlow 2 中重写的代码，其中包含关键更改和说明：

1。导入必要的图层：

from tensorflow.keras.layers import LSTMCell, RNN, Dropout

2。创建 LSTM 单元

lstm_cells = [LSTMCell(size_layer) for _ in range(num_layers)]

3.构建带 dropout 的 RNN 层：

rnn = RNN(
    lstm_cells,
    return_sequences=True,  # Set to True if you need to return all outputs
    return_state=False,    # Set to True if you need to return the final state
)

drop = Dropout(forget_bias)(rnn)  # Apply dropout after the RNN layer