Function main

evaluation.py:183–289

()

                pass

def main():
    parser = argparse.ArgumentParser()
    parser.add_argument('--text_file', type=str)
    parser.add_argument('--pred_file', type=str)
    parser.add_argument('--gold_file', type=str)
    parser.add_argument('--schema_file', type=str)

    parser.add_argument('--format', type=str, default="dyiepp")
    parser.add_argument('--verbose', action='store_true')
    parser.add_argument('--decoding_format', type=str, default='noetrtspan')
    options = parser.parse_args()


    label_schema = EventSchema.read_from_file(
        filename=options.schema_file
    )

    decoding_format_dict = {
        'role': RolePredictParser,
        'tri': TriPredictParser
    }

    # 替换为自己的predict parser
    pred_reader = decoding_format_dict[options.decoding_format](schema=label_schema) 


    trigger_metric = Metric()
    argument_metric = Metric()

    # Reconstruct the offset of predicted event records.
    text_filename = options.text_file
    pred_filename = options.pred_file
    gold_filename = options.gold_file
    print("pred_filename: ", pred_filename)
    print("gold_filename: ", gold_filename)
    
    # 离线评估
    # 在此处处理的时候, 需要将 et、rt特殊处理的部分进行添加, 以及src相同的部分进行合并
    event_list, _ = pred_reader.decode(
        gold_list=[],
        pred_list=read_file(pred_filename),
        text_list=[json.loads(line)['text']
                    for line in read_file(text_filename)],
    )
    # print(event_list[0])

    # text 中空格一类的做key会有影响, 后续可考虑用id来指代
    text_pred_dict = {} # 构建 text: ([tri_list][role_list]) 类型的字典 
    text_gold_dict = {}
    
    for item in event_list:
        if item["text"] in text_pred_dict:
            # print("Warning: text duplicate , text: ", item["text"])
            text_pred_dict[item["text"]][0] += item['pred_event']
            text_pred_dict[item["text"]][1] += item['pred_role']
        else:
            text_pred_dict[item["text"]] = [item['pred_event'], item['pred_role']]