知识卡片
排名指标不自动惩罚概率失准
严格单调变换可保持ROC-AUC而改变数值概率与校准。因此不能用AUC单独证明置信度可靠或自信错误惩罚充分;概率质量可另看适当损失与校准,分类行动还需阈值及误差成本。
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
The ROC AUC micro average score is essentially the sum of all the predict_proba error values, or how far the predicted probability values are from the 0 or 1 values that each example was given by a human labeler. It’s always a good idea to have that mental model in mind when you are measuring model accuracy. Accuracy is just how close to what your
参考来源
- 位置:《Natural Language Processing in Action, Second Edition》第09章《可堆叠的Transformer》(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-9.html)
- 结论依据:摘录中关于排名指标不自动惩罚概率失准的定义、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:The ROC AUC micro average score is essentially the sum of all the predict_proba error values, or how far the predicted probability values are from the 0 or 1 values that each example was given by a human labeler. It’s always a good idea to have that mental model in mind when you are measuring model accuracy. Accuracy is just how close to what your