TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Are polynomial features the root of all evil? A myth

1 点作者 hardmaru超过 1 年前

1 comment

bell-cot超过 1 年前
First two sentences: &quot;When fitting a non-linear model using linear regression, we typically generate new features using non-linear functions. We also know that any function, in theory, can be approximated by a sufficiently high degree polynomial.&quot;<p>Reaction after one glance at the graph right below the first paragraph: If the data points you are trying to fit look <i>that</i> sinusoidal, why the [bleep&#x2F;] are you trying to fit them with polynomials, and not trig functions?<p>A couples sentences later: &quot;But many blogs, papers, and even books tell us that high polynomials should be avoided. They tend to oscilate [sic] and overfit!&quot;<p>Reaction: sounds to me like many of the authors in this field need to take a decent math and stats class or few - to learn about power series, and about the too-many-parameter problem, and about test-fitting curves to subsets of your data points to test the robustness of your analysis, and about the dangers of idiotic extrapolation from limited data.<p>Lazy meta-reaction: The article has been on HN for 40 minute now, and gotten zero upvotes. I&#x27;ll hope that means that everyone here knows basic math &amp; data analysis, and move on.