backup/utils/prepare_test_submission_file.py

import json
import csv
if __name__ == '__main__':
    # category_to_sample_ids_dict = json.load(open('../category_to_sample_ids_dict.json'))
    # answer_dict_for_mixtral_eval = json.load(open('../answer_dict_for_mixtral_eval.json'))

    model_name = 'llava1.5_7b'
    # model_name = 'llava1.5_13b'

    submission_file_phase1_6datasets = f'/system/user/publicdata/LMM_benchmarks/MMFM_challenge_final/doc-vl-eval/inference_results/output_6_datasets_{model_name}.json'
    submission_file_phase1_4datasets = f'/system/user/publicdata/LMM_benchmarks/MMFM_challenge_final/doc-vl-eval/inference_results/output_4_datasets_{model_name}.json'
    submission_file_phase2 = f'/system/user/publicdata/LMM_benchmarks/MMFM_challenge_final/doc-vl-eval/inference_results/output_private_{model_name}.json'
    submission_file = f'test_submission_{model_name}.csv'


    # answer_dict_file = '/system/user/publicdata/LMM_benchmarks/MMFM_challenge_final/doc-vl-eval/answer_dicts/coverted_output_test_all_datasets.json'
    answer_dict_file_phase1 = '/system/user/publicdata/LMM_benchmarks/MMFM_challenge_final/doc-vl-eval/answer_dicts/coverted_output_test_all_datasets.json'
    answer_dict_file_phase2 = '/system/user/publicdata/LMM_benchmarks/MMFM_challenge_final/doc-vl-eval/answer_dicts/converted_output_test_private_for_mixtral_eval.json'

    with open(answer_dict_file_phase1, 'r') as f:
        answer_dict_phase1 = json.load(f)  # 2000 samples,  200 samples x 10 datasets
    with open(answer_dict_file_phase2, 'r') as f:
        answer_dict_phase2 = json.load(f)


    submission_phase1_6datasets = json.load(open(submission_file_phase1_6datasets))
    submission_phase1_4datasets = json.load(open(submission_file_phase1_4datasets))
    submission_phase2 = json.load(open(submission_file_phase2))

    answer_dict = {**answer_dict_phase1, **answer_dict_phase2}
    merged_submission = {**submission_phase1_6datasets, **submission_phase1_4datasets, **submission_phase2}

    # answer_dict = { **answer_dict_phase2}
    # merged_submission = {**submission_phase2}

    merged_submission_ = {}
    for key, value in merged_submission.items():
        items = key.split('_')
        if items[0] == 'mychart':
            items = items[:3] + [items[-1]]
        elif items[0] == 'myinfographic':
            items = items[:2] + [items[-1]]
        key_ = '_'.join(items[:-1])
        merged_submission_[key_] = value

    with open(submission_file, 'w', newline='') as f:
        writer = csv.writer(f)
        writer.writerow(['id', 'pred', 'split'])
        # idx = 0
        for sample_key, data_dict in answer_dict.items():
            items = sample_key.split('_')
            id_ = '_'.join(items[:-1])
            pred = merged_submission_[id_]
            writer.writerow([id_, pred, 'public'])
            # idx += 1

    t = 1