sean0042 / open_wikitable Goto Github PK

Dear authors,

First of all, thanks for your work which is helpful and inspiring. I found your dataset and code repo very useful.

I see in your data folder that you provide the already flattened tables in chunks. May I ask if it is possible that you also share the code that parses the raw table into the chunked and flattened format? That will be very helpful for our research.

Thanks a lot.

I find there are several fields of each sample in the dataset, which are dict_keys(['question_id', 'original_table_id', 'question', 'sql', 'answer', 'hard_positive_idx', 'positive_idx', 'negative_idx', 'dataset']).

1. Could you provide a comment about the meaning of each field?
2. Literally, positive idx means the index of relevant tables, however, I find some positive_idx are empty lists ([]), does it mean there are no relevant tables for this question？ and how do you feed empty tables into the encoder?