backup/utils/prepare_sample_answer_dict_files.py

import json
import csv
if __name__ == '__main__':
    answer_dict_file = '/system/user/publicdata/LMM_benchmarks/MMFM_challenge_final/doc-vl-eval/answer_dicts/coverted_output_test_all_datasets.json'
    answer_dict_w_question_phase1_file = '/system/user/publicdata/LMM_benchmarks/MMFM_challenge_final/doc-vl-eval/answer_dicts/converted_output_test_for_mixtral_eval.json'
    answer_dict_w_question_phase2_file = '/system/user/publicdata/LMM_benchmarks/MMFM_challenge_final/doc-vl-eval/answer_dicts/converted_output_test_private_for_mixtral_eval.json'
    solution_file = 'MMFM_Challenge_solution.csv'
    category_to_sample_ids_dict_file = 'category_to_sample_ids_dict.json'
    answer_dict_for_mixtral_eval_file = 'answer_dict_for_mixtral_eval.json'

    header = ['id', 'pred', 'split']
    # id  is in format docvqa_docvqa_0_4

    # answer_dict = {}
    with open(answer_dict_file, 'r') as f:
        answer_dict = json.load(f)  # 2000 samples,  200 samples x 10 datasets
    answer_dict_w_question_phase1 = json.load(open(answer_dict_w_question_phase1_file))  # 800(=200*4) samples
    answer_dict_w_question_phase2 = json.load(open(answer_dict_w_question_phase2_file))  # 200+400+428=1028 samples



    answer_dict = {**answer_dict, **answer_dict_w_question_phase2}
    answer_dict_w_question = {**answer_dict_w_question_phase1, **answer_dict_w_question_phase2}

    category_to_sample_ids_dict = {}
    answer_dict_for_mixtral_eval = {}

    datasets_w_mixtral_eval_phase1 = ['docvqa', 'infographicvqa', 'websrc', 'wtq']
    datasets_w_mixtral_eval_phase2 = ['mydoc', 'mychart', 'myinfographic']


    long_key_to_short_key_in_answer_dict = {}
    short_key_to_long_key_in_answer_dict = {}
    for sample_key in answer_dict.keys():
        short_key = '_'.join( sample_key.split('_')[:-1])
        long_key_to_short_key_in_answer_dict.update({sample_key: short_key})
        short_key_to_long_key_in_answer_dict.update({short_key: sample_key})

    long_key_to_short_key_in_answer_dict_w_question = {}
    short_key_to_long_key_in_answer_dict_w_question = {}
    for sample_key in answer_dict_w_question.keys():
        short_key = '_'.join( sample_key.split('_')[:-1])
        long_key_to_short_key_in_answer_dict_w_question.update({sample_key: short_key})
        short_key_to_long_key_in_answer_dict_w_question.update({short_key: sample_key})


    with open(solution_file, 'w', newline='') as f:
        writer = csv.writer(f)
        writer.writerow(header)
        dataset_counter_dict = {}
        for sample_key, data_dict in answer_dict.items():
            items = sample_key.split('_')
            id_ = '_'.join(items[:-1])
            if items[0] in datasets_w_mixtral_eval_phase2:
                category = items[0]
            else:
                if items[0] == items[1]:
                    category = items[0]
                else:
                    category = items[0] + '_' + items[1]
            if category not in category_to_sample_ids_dict:
                category_to_sample_ids_dict[category] = []
            if category not in dataset_counter_dict:
                dataset_counter_dict[category] = 0

            # id_ = f'{category}_{dataset_counter_dict[category]}'
            category_to_sample_ids_dict[category].append(id_)
            # write to the solution file
            writer.writerow([id_, data_dict['ground_truth'], 'public'])

            if category in datasets_w_mixtral_eval_phase1 + datasets_w_mixtral_eval_phase2:
                short_key = long_key_to_short_key_in_answer_dict[sample_key]
                sample_key_w_question = short_key_to_long_key_in_answer_dict_w_question[short_key]
                answer_dict_for_mixtral_eval.update({
                    id_: {
                        "question_type": "short-answer",
                        "ground_truth": answer_dict_w_question[sample_key_w_question]['ground_truth'],
                        'question': answer_dict_w_question[sample_key_w_question]['question'],
                        "dataset": category,
                    }
                })

            dataset_counter_dict[category] += 1
    with open(category_to_sample_ids_dict_file, 'w') as f:
        json.dump(category_to_sample_ids_dict, f)
    with open(answer_dict_for_mixtral_eval_file, 'w') as f:
        json.dump(answer_dict_for_mixtral_eval, f)