知识卡片
类别均衡不保证每类同样准确
均衡采样改变训练目标对各类的关注,常有助避免多数类主导,但难度、样本质量和重叠仍不同。测试还须反映目标用途或分别报告每类指标,不能由样本数量相等推断预测准确率已公平。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
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 will be using the deduplicated dataset you loaded in listing 8.5. To save you the trouble of downloading a bloated dataset, we’ve provided the statistics on duplicates, and you will use a balanced sampling of countries to encourage your model to treat all categories and names equally. This means your model will predict common names and countries just as accurately as popular names from
参考来源
- 位置:《Natural Language Processing in Action, Second Edition》第08章《循环网络与长短期记忆》(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-8.html)
- 结论依据:摘录中关于类别均衡不保证每类同样准确的定义、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 will be using the deduplicated dataset you loaded in listing 8.5. To save you the trouble of downloading a bloated dataset, we’ve provided the statistics on duplicates, and you will use a balanced sampling of countries to encourage your model to treat all categories and names equally. This means your model will predict common names and countries just as accurately as popular names from