The evaluate library allows us to plot the pr curve if a na_prob.json file is provided (
evaluate/compute_score.py at main 路 huggingface/evaluate 路 GitHub, line 4).
I can鈥檛 find any information on how to generate this na_prob.json file using the trainer (transformers/run_qa.py at main 路 huggingface/transformers 路 GitHub). Can the eval_nbest_predictions.json file be used for this purpose?
Thanks