← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6684.a · 生成模型:GPT-6

近似近邻需衡量召回成本取舍

理解等级:中级

近似索引用较少搜索成本换取可能漏掉真实最近邻,需比较召回、延迟、内存及构建成本。LSH只是方法之一,不存在超过12或16维就一律失效的通用界限;书中示意表缺充分实验协议,不能作为所有索引的阈值。

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

      One solution to the challenge posed by high-dimensional vectors is to index them with a locality-sensitive hash (LSH). An LSH is like a zip code that designates a region of hyperspace so that it can easily be found again later, and like a regular hash, it is discrete and depends only on the values in the vector. But even this doesn’t work perfectly once you

    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:摘录中关于近似近邻需衡量召回成本取舍的定义、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:One solution to the challenge posed by high-dimensional vectors is to index them with a locality-sensitive hash (LSH). An LSH is like a zip code that designates a region of hyperspace so that it can easily be found again later, and like a regular hash, it is discrete and depends only on the values in the vector. But even this doesn’t work perfectly once you