JAQKET_baseline

Home: https://www.nlp.ecei.tohoku.ac.jp/projects/aio/

LeaderBoard: https://www.nlp.ecei.tohoku.ac.jp/projects/AIP-LB/task/aio

Requirements

python 3.6 or later

Several python liblaries

$ pip install tqdm
$ pip install numpy
$ pip install torch
$ pip install tensorboard
$ pip install transformers
$ pip install mecab-python3

Sample

$ mkdir ../data
$ cd ../data
$ wget https://jaqket.s3.ap-northeast-1.amazonaws.com/data/aio_01/train_questions.json
$ wget https://jaqket.s3.ap-northeast-1.amazonaws.com/data/aio_01/dev1_questions.json
$ wget https://jaqket.s3.ap-northeast-1.amazonaws.com/data/aio_01/dev2_questions.json
$ wget https://jaqket.s3.ap-northeast-1.amazonaws.com/data/aio_01/all_entities.json.gz

$ mkdir ../working_dir
$ cd ../working_dir

$ cp ../JAQKET_baseline/run_jaqket_baseline_sample.sh .

$ ./run_jaqket_baseline_sample.sh

Launch

You can run these setting commands using ./scripts/preprocess.sh

submit

After training, you would convert a prediction file into a submission format.

$ cd [your working directory]
$ python ../JAQKET_baseline/scripts/create_submission_file.py \
    -test ../data/aio_leaderboard.json \
    -pred output_dir/is_test_true_output_labels.txt \
    -fo submission.json \
    -wq -wc

And that, try to submit submission.json from the following page: https://www.nlp.ecei.tohoku.ac.jp/projects/AIP-LB/task/aio