← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6681.d · 生成模型:GPT-6

主题数是需要验证的超参数

理解等级:中级

主题数影响粒度、估计稳定性、成本和可解释性,无法仅靠先验就自动得到唯一正确答案。可按留出数据、主题解释或下游任务选择,但需明确目标;k近邻中的k是邻居数,不是与k均值相同的主题或簇数。

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 second parameter you need to specify for an LDiA model, the number of topics, is a bit trickier. The number of topics in a particular set of documents can’t be measured directly until after you’ve assigned words to those topics. Like k-means, k-nearest neighbor, and other clustering algorithms, you must tell it the k ahead of time. You can guess the number of topics

    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:摘录中关于主题数是需要验证的超参数的定义、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 second parameter you need to specify for an LDiA model, the number of topics, is a bit trickier. The number of topics in a particular set of documents can’t be measured directly until after you’ve assigned words to those topics. Like k-means, k-nearest neighbor, and other clustering algorithms, you must tell it the k ahead of time. You can guess the number of topics