gen_qa

Note:

The code is based on Huggingface transformer examples: question-answering

Instruction:

Data preprocess:

conda activate torch_transformer
cd preprocess
python produce_genqa_stage1_data.py
python produce_genda_stage2_gold.py
cp -r ../data/gen_qa/* ../gen_qa/data/

Train stage1 model:

conda activate torch_transformer
cd gen_qa
python run_seq2seq_qa.py training_arguments_stage1.json

Evaluate stage1 model event (type) classification F1 (optional):

python eval_stage1_event.py

Transfer stage1 output to stage2 input:

python transfer_s1_pred_to_s2_input.py

Train stage2 model:

python run_seq2seq_qa.py training_arguments_stage2.json

Evaluate stage2 on test set:

python eval_stage2_preds.py