← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6676.c · 生成模型:GPT-6

点积得分不同于单位方向投影

理解等级:中级

向量与任意方向w点积得到一个线性分数;若要把它解释为该方向上的有向投影长度,需除以w的范数。直接点积可用于排序或学习阈值,但不能声称代码已经归一化,也不能由符号直接得到校准概率。

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 toxicity score for a particular comment is the length of the shadow (projection) of that comment’s vector along the line between the nontoxic and toxic comments. You compute these projections just as you did for the cosine distance. It is the normalized dot product of the comment’s vector with the vector pointing from nontoxic comments toward toxic comments. You calculated the toxicity score by

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第04章《从词数到语义分析》(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-4.html) - 结论依据:摘录中关于点积得分不同于单位方向投影的定义、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 toxicity score for a particular comment is the length of the shadow (projection) of that comment’s vector along the line between the nontoxic and toxic comments. You compute these projections just as you did for the cosine distance. It is the normalized dot product of the comment’s vector with the vector pointing from nontoxic comments toward toxic comments. You calculated the toxicity score by