恢复模型

恢复也很好，很容易。

这是一个方便的帮助函数：

def restore_vars(saver, sess, chkpt_dir):
    """ Restore saved net, global score and step, and epsilons OR
    create checkpoint directory for later storage. """
    sess.run(tf.initialize_all_variables())

    checkpoint_dir = chkpt_dir 

    if not os.path.exists(checkpoint_dir):
        try:
            print("making checkpoint_dir")
            os.makedirs(checkpoint_dir)
            return False
        except OSError:
            raise

    path = tf.train.get_checkpoint_state(checkpoint_dir)
    print("path = ",path)
    if path is None:
        return False
    else:
        saver.restore(sess, path.model_checkpoint_path)
        return True

主要代码：

path_to_saved_model = './'
max_steps = 1

# Start a session
with tf.Session() as sess:

    ... define the model here ...

    print("define the param saver")
    saver = tf.train.Saver(max_to_keep=5, keep_checkpoint_every_n_hours=1)

    # restore session if there is a saved checkpoint
    print("restoring model")
    restored = restore_vars(saver, sess, path_to_saved_model)
    print("model restored ",restored)

    # Now continue training if you so choose

    for step in range(max_steps):

        # do an update on the model (not needed)
        loss_value = sess.run([loss])
        # Now save the model
        saver.save(sess, "./model", global_step=step)