skywalker518 / a-recsys Goto Github PK

Hello, I found a performance issue in the definition of _var_indices, attributes/embed_attribute.py, tf.zeros will be created repeatedly during program execution, resulting in reduced efficiency. I think it should be created before the loop .

Looking forward to your reply. Btw, I am very glad to create a PR to fix it if you are too busy.

I use tensorflow r0.12 to run ml-1m dataset and i get the following error.

Traceback (most recent call last):
File "/home/endy/PycharmProjects/recommender/lstm/run.py", line 720, in
main()
File "/home/endy/PycharmProjects/recommender/lstm/run.py", line 717, in main
train()
File "/home/endy/PycharmProjects/recommender/lstm/run.py", line 360, in train
train_set, dev_set, test_set, embAttr, START_ID, item_population, p_item, _, _ = read_data(task)
File "/home/endy/PycharmProjects/recommender/lstm/run.py", line 299, in read_data
embAttr = embed_attribute.EmbeddingAttribute(u_attr, i_attr, FLAGS.batch_size, FLAGS.n_sampled, _buckets[-1], FLAGS.use_sep_item, item_ind2logit_ind, logit_ind2item_ind, devices=devices)
File "/home/endy/PycharmProjects/recommender/lstm/attributes/embed_attribute.py", line 116, in init
self.update_sampled = self._pass_sampled_items(prefix, device=devices[-1])
File "/home/endy/PycharmProjects/recommender/lstm/attributes/embed_attribute.py", line 337, in _pass_sampled_items
mulhot_indices = batch_slice2(self.att[prefix][1][i], b, s, self.n_sampled)
File "/home/endy/PycharmProjects/recommender/lstm/attributes/mulhot_index.py", line 46, in batch_slice2
return tf.concat(res, 0)
File "/home/endy/.virtualenvs/tensorflow/local/lib/python2.7/site-packages/tensorflow/python/ops/array_ops.py", line 1122, in concat
).assert_is_compatible_with(tensor_shape.scalar())
File "/home/endy/.virtualenvs/tensorflow/local/lib/python2.7/site-packages/tensorflow/python/framework/tensor_shape.py", line 756, in assert_is_compatible_with
raise ValueError("Shapes %s and %s are incompatible" % (self, other))
ValueError: Shapes (1024, ?) and () are incompatible