← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6693.a · 生成模型:GPT-6

样本权重改变损失代表的总体

理解等级:中级

用聚合记录count加权,可让高频记录对目标贡献更多,区别于每个姓名标签行同权。默认mean一般按元素数除,而非自动除以权重和;损失量级随平均权重变化,比较训练测试时需说明归约口径。

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

      The logarithmic nature of the function allows it to penalize a “confidently wrong” example, when your model predicts with high probability that the sex of a particular name is male, when it is actually more commonly labeled as female. We can help it make the penalties even more related to reality by using another piece of information available to us—the frequency of the name for

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第05章《神经网络基础》(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-5.html) - 结论依据:摘录中关于样本权重改变损失代表的总体的定义、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 logarithmic nature of the function allows it to penalize a “confidently wrong” example, when your model predicts with high probability that the sex of a particular name is male, when it is actually more commonly labeled as female. We can help it make the penalties even more related to reality by using another piece of information available to us—the frequency of the name for