search

coastalcph / seq2sparql

Multilingual Compositional Wikidata Questions (MCWQ)
18 stars 0 forks source link

Input file path in hf/run_mt5.sh #5

Closed mihaela-bornea closed 2 years ago

mihaela-bornea commented 2 years ago

This is a very interesting dataset and the baseline and evaluation code are very helpful.

I have a clarification question as the input path from the run scripts does not match exactly the structure of the data directory in your git repo.

Please confirm if parameter values from hf/run_mt5.sh:

--train_file ../data/translations/$split/train.$lang.json \
--validation_file ../data/translations/$split/dev.$lang.json \
--test_file ../data/translations/$split/test.$lang.json \

should point to 

../mcwq/translations/$split/train.$lang.sparql
../mcwq/translations/$split/train.$lang.txt

../mcwq/translations/$split/dev.$lang.sparql
../mcwq/translations/$split/dev.$lang.txt

../mcwq/translations/$split/test.$lang.sparql
../mcwq/translations/$split/test.$lang.txt
ruixiangcui commented 2 years ago

Hi Mihaela @mihaela-bornea , thanks for your interest and for pointing out the mistake! The JSON files are a concatenation of the question and query pairs. I have uploaded the files and the script. Please pull from the latest commit.