访问 RNN 权重 - Tensorflow

Question

我正在使用tf.python.ops.rnn_cell.GRUCell

output, state = tf.nn.dynamic_rnn(
        GRUCell(HID_DIM),
        sequence,
        dtype=tf.float32,
        sequence_length=length(sequence)
)

如何获取这个 GRUCell 的权重。我需要查看它们以进行调试。

Answer 1

可以使用以下方式打印当前会话中所有变量的值：

with tf.Session() as sess:
    variables_names =[v.name for v in tf.trainable_variables()]
    values = sess.run(variables_names)
    for k,v in zip(variables_names, values):
        print(k, v)

访问 RNN 权重 - Tensorflow

Accessing RNN weights- Tensorflow

python

lstm

tensorflow

recurrent-neural-network