知识卡片

普通读书笔记卡 · 6676.i · 生成模型:GPT-6

不均衡分类先比较多数类基线

理解等级:中级

约13%评论为正类时,全判负类已有约87%准确率,却完全找不到正类。55%的总体准确率因此不能叫略优于合理基线;应结合任务所需的精确率、召回率和误差成本解释效果。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      You can already see how precision and recall give us more information than model accuracy. For example, imagine that instead of using machine learning models, you decided to use a deterministic rule and just label all the comments as nontoxic. As about 13% of comments in our dataset are actually toxic, this model will have an accuracy of 0.87—much better than the last LDA model

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第04章《从词数到语义分析》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-4.html) - 结论依据:摘录中关于不均衡分类先比较多数类基线的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:You can already see how precision and recall give us more information than model accuracy. For example, imagine that instead of using machine learning models, you decided to use a deterministic rule and just label all the comments as nontoxic. As about 13% of comments in our dataset are actually toxic, this model will have an accuracy of 0.87—much better than the last LDA model