[ASK] Unexpected Behavior with LightGCN Model When Loading Saved Files

[Henry Wu]

Description

I am currently using the LightGCN model for predictions on my own dataset. The regular workflow, including initializing the model, model.fit(), and model.recommend_k_items(), functions without any issues.

However, after I use model.load() to restore the model from previously saved data, model.recommend_k_items() throws an error:

---------------------------------------------------------------------------
InvalidArgumentError                      Traceback (most recent call last)
File ~\AppData\Roaming\Python\Python39\site-packages\tensorflow\python\client\session.py:1402, in BaseSession._do_call(self, fn, *args)
   1401 try:
-> 1402   return fn(*args)
   1403 except errors.OpError as e:

File ~\AppData\Roaming\Python\Python39\site-packages\tensorflow\python\client\session.py:1385, in BaseSession._do_run.<locals>._run_fn(feed_dict, fetch_list, target_list, options, run_metadata)
   1384 self._extend_graph()
-> 1385 return self._call_tf_sessionrun(options, feed_dict, fetch_list,
   1386                                 target_list, run_metadata)

File ~\AppData\Roaming\Python\Python39\site-packages\tensorflow\python\client\session.py:1478, in BaseSession._call_tf_sessionrun(self, options, feed_dict, fetch_list, target_list, run_metadata)
   1476 def _call_tf_sessionrun(self, options, feed_dict, fetch_list, target_list,
   1477                         run_metadata):
-> 1478   return tf_session.TF_SessionRun_wrapper(self._session, options, feed_dict,
   1479                                           fetch_list, target_list,
   1480                                           run_metadata)

InvalidArgumentError: Cannot multiply A and B because inner dimension does not match: 2160 vs. 2162.  Did you forget a transpose?  Dimensions of A: [2160, 2160).  Dimensions of B: [2162,64]
	 [[{{node SparseTensorDenseMatMul/SparseTensorDenseMatMul}}]]

Please assist in resolving this issue. Thank you!

Other Comments

[Henry Wu]

Update: Ignore this, it does not work at all.

For those who encounter this issue, you have to restore the model like this:

model.saver = tf.compat.v1.train.import_meta_graph('./model/lightgcn.meta')
model.saver.restore(model.sess, tf.train.latest_checkpoint('./model'))