为什么执行了python3 main.py --do_train --do_eval后result.txt文件是空的？

[yangxia605]

你好：
请教一下，为什么执行了python3 main.py --do_train --do_eval之后，result.txt 文件是空的，请问这是什么原因呢？谢谢

[Pvvv]

你好：
请教一下，为什么执行了python3 main.py --do_train --do_eval之后，result.txt 文件是空的，请问这是什么原因呢？谢谢

请问您的问题现在解决了嘛

[hialoha]

I also have the same problem,result.txt is empty,and I haven't deal with it.

[Claymore715]

我也有这个问题

[Pvvv]

得下载一个 perl   才可以

…

----------------- 原始邮件 ------------------ 发件人: "monologg/R-BERT" ***@***.***>; 发送时间: 2021年5月20日(星期四) 晚上9:27 ***@***.***>; ***@***.******@***.***>; 主题: Re: [monologg/R-BERT] 为什么执行了python3 main.py --do_train --do_eval后result.txt文件是空的？ (#12) 我也有这个问题 — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

[wjx-git]

除了安装Perl外，还需要将 eval/answer_keys.txt 中的标签改为你自己验证集数据的标签。还需要更改 semeval2010_task8_scorer-v1.2.pl 中的部分内容，挺麻烦的，还不如用 sklearn 计算。

[lixianglong1205]

使用sklearn计算f1-score
欢迎指正

import os
from sklearn.metrics import f1_score, precision_score, accuracy_score, recall_score

EVAL_DIR = "eval"


def txt_result_analysis(path):
    with open(path, mode="r", encoding="utf-8") as answer_file:
        source_label = [i.rstrip() for i in answer_file if i.strip()]
    label_dict = {}
    for one in source_label:
        i, j = one.split("\t")
        label_dict[i] = j

    return label_dict


def official_f1(average="macro"):
    """使用sklearn包计算f1-score"""
    source_label_dict = txt_result_analysis("./eval/answer_keys.txt")
    proposed_label_dict = txt_result_analysis("./eval/proposed_answers.txt")
    source_label, proposed_label = [], []
    for key in source_label_dict:
        value1 = source_label_dict[key]
        value2 = proposed_label_dict[key]
        source_label.append(value1)
        proposed_label.append(value2)
    """
    micro: 计算所有的TP TN FP FN，然后计算F1-score
    f1_score_ = f1_score(source_label, proposed_label, average="micro")
    macro: 计算每一个类别的F1-score，然后对所有的F1-scre求平均值
    f1_score_ = f1_score(source_label, proposed_label, average="macro")
    """
    f1_score_ = f1_score(source_label, proposed_label, average=average)
    return f1_score_


if __name__ == "__main__":
    print("macro-averaged F1 = {}%".format(official_f1() * 100))