知识卡片
零一错误不能直接当连续概率
统计误分类前需将连续分数按argmax或阈值转为离散决策。直接比较概率数值是否等于0/1会把绝大多数正常预测算错;式50.6的一位位不相等也可能把一次单标签错误计两次,需说明统计单位。
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
- · 目标识别
原书出处 · 1 处
- · 查看原书依据
In order to learn the model parameters, we need to define a loss function that will capture the task that we want to solve. One natural measure of classification error is the missclassication error. The errors that the function f makes are the number of misplaced ones (i.e., the misclassification error) over the dataset, which we can write as:
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第50章《目标识别》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)/_epub-src/OPS/xhtml/c050b.xhtml)
- 结论依据:摘录中关于零一错误不能直接当连续概率的定义、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:In order to learn the model parameters, we need to define a loss function that will capture the task that we want to solve. One natural measure of classification error is the missclassication error. The errors that the function f makes are the number of misplaced ones (i.e., the misclassification error) over the dataset, which we can write as: