如何在TensorFlow中恢复检查点目录？

Question

I train model with method in TensorFlow tutorials (code is here ). At last I save the model in a checkpoint directory. Now I want to restore from the checkpoint directory:

import tensorflow as tf

def main(_):
    saver = tf.train.Saver()
    with tf.Session() as sess:
        ckpt = tf.train.latest_checkpoint("/data/lstm_models")
        saver.restore(sess, ckpt)

if __name__ == "__main__":
  tf.app.run()

Howerver, I got error:

ValueError: No variables to save

Answer 1

It look like you haven't defined your graph to restore from the checkpoint, so when building the saver it complains that your graph is empty.

Could you try to build your graph again (eg redefining your variables) before trying to restore it?

From the restore method string doc:

It requires a session in which the graph was launched.