Tensorflow模型恢复问题

Question

我在Tensorflow中的RNN模型恢复期间一直发现以下问题。该模型运行良好，但恢复有问题。下面是我的代码。可能是什么问题。我删除并清除了之前的所有会话，但同样的问题仍然存在。

saver = tf.train.import_meta_graph(os.path.join(logs_path, 
'Multi_Layer_RNN_model.ckpt-499999.meta'))

with tf.Session() as sess:
    ckpt = tf.train.get_checkpoint_state(logs_path)
    saver.restore(sess, tf.train.latest_checkpoint(logs_path))  

错误

InvalidArgumentError: Assign requires shapes of both tensors to match. lhs shape= [105,400] rhs shape= [111,400]
     [[Node: save/Assign_11 = Assign[T=DT_FLOAT, _class=["loc:@rnn/lstm_cell/kernel"], use_locking=true, validate_shape=true, _device="/job:localhost/replica:0/task:0/device:CPU:0"](rnn/lstm_cell/kernel/RMSProp_1, save/RestoreV2:11)]]

Caused by op 'save/Assign_11', defined at:
  File "C:\Users\ADMN\AppData\Local\conda\conda\envs\tensorflow\lib\runpy.py", line 193, in _run_module_as_main
    "__main__", mod_spec)
  File "C:\Users\ADMN\AppData\Local\conda\conda\envs\tensorflow\lib\runpy.py", line 85, in _run_code
    exec(code, run_globals)
  File "C:\Users\ADMN\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\ipykernel_launcher.py", line 16, in <module>
    app.launch_new_instance()
  File "C:\Users\ADMN\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\traitlets\config\application.py", line 658, in launch_instance
    app.start()
  .................

Answer 1

在第一行你有理由：

lhs shape= [105,400] rhs shape= [111,400]

看起来你在神经网络之间有不同的向量/张量维度。