知识卡片
阈值提高不保证精确率上升
提高固定分数阈值会减少预测正类,召回通常不增,但精确率是否提高还取决于被去掉样本的标签。置信分数不自动校准,0.5也不总最优;应按验证数据和成本选阈值。
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
When you try to solve any of the classification problems we deal with in this book, you might notice that your classification model rarely returns a discrete label, such as true or false. More often, it returns a number between 0 and 1—the probability of the given record to belong to the predicted class. In the world of intent recognition, which we cover extensively in
参考来源
- 位置:《Natural Language Processing in Action, Second Edition》附录D《附录D:机器学习工具与技术》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/appendix-d.html)
- 结论依据:摘录中关于阈值提高不保证精确率上升的定义、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:When you try to solve any of the classification problems we deal with in this book, you might notice that your classification model rarely returns a discrete label, such as true or false. More often, it returns a number between 0 and 1—the probability of the given record to belong to the predicted class. In the world of intent recognition, which we cover extensively in