hyren / query2box Goto Github PK

Thanks for opening code of such a good work first!

I notice in the paper the MRR of model on FB15k-237 is 0.295 (Table 4), and the MRR on FB15k-237 about 1p is 0.4 (Table 9). I want to know what's the difference between these two results (from my understanding they are the same structure of data so the model should reach the identical performance? )

Thanks a lot again.

Excuse me, how is the data of query built? For example, the file train-queries.pkl.

May I know how to convert natural language questions into logical queries? Like the example ((8610, (146, 1093), 0), '2-chain') in test_triples_2c.pkl.
Is it possible that you share data processing code?
Thanks!

Hello Professor, the program runs to the half to report the error as shown in the figure

Hi Ren:

May I know how you build the dataset?
Is it possible that you release the original dataset and data processing code?

Hi
Thanks for sharing the code, how I can generate files like trian_triples_1c.pkl, trian_triples_2c.pkl.
Is it possible that you release the data processing code?
Thanks！

Hi Ren:

I was wondering about the evaluation/answer sets used for obtaining the designated metrics, i.e. MRR, Hits@K .

There are 2 pickled answer dictionaries for each type of chain, i.e. "test_ans_ic.pkl and test_ans_1c_hard.pkl"

According to

Only the "_hard" version of answers are used for evaluation. I wanted to clarify the meaning and origin of the "_hard" answers as I couldn't find it in the paper.

It seems that the normal answers are only used for find "false_answers" which in turn are used to filter the scores in

If possible can you elaborate a bit on this chunk ?

It would be really useful if this code were pip-installable to make its installation and usage more reproducible. Would you be willing to accept a PR for this?

Thank you for providing the implementation for query2box.

I have discovered some discrepancies between the implementation of "ip" "pi", and "ui" queries and what is described in the paper, mainly regarding mismatch of the number of projections in these three queries types.

For "pi", the code is implemented like:

there are two projections from set 1 and one projection from set 2.
However, according to the paper, the "pi" should be one projection from set 1 and no projection conducted for set 2.

A similar case is also found at "pi" and "ui", in which there exist extra projections in code than what is described in the paper.

I was wondering whether there is some misunderstanding of the code or I have misinterpreted the results showed in the paper.

The pkl files in the data directory only have the integer ids of the entities and relations
Could you release the mapping from these ids to entity and relation names?