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。

Everything is a linear model

224 点作者 nopipeline超过 1 年前

16 条评论

levocardia超过 1 年前
I see a lot of comments here assuming &quot;linear model&quot; means &quot;can&#x27;t model nonlinearities.&quot; Absolutely not the case. Splines can easily take care of that. The &quot;linear&quot; part of linear model just means &quot;linear in the predictor space.&quot; You can add a non-linear predictor easily via spline basis (similar&#x2F;sometimes identical to &quot;kernels&quot; in ML).<p>My series of lm&#x2F;glm&#x2F;gam&#x2F;gamm revelations was:<p>1. All t-tests and ANOVA flavors are just linear models<p>2. Linear models are just a special case of generalized linear models (GLMs), which can deal with binary or count data too<p>3. All linear models and GLMs are just special cases of generalized linear <i>mixed</i> models, which can deal with repeated measures, grouped data, and other non-iid clustering<p>4. Linearity is usually a bad assumption, which can easily be overcome via splines<p>5. Estimating a spline smoothing penalty is the same thing as estimating the variance for a random effect in a a mixed model, so #3 and #4 can be combined for free<p>And then you end up with a generalized additive mixed model (GAMM), which can model smooth nonlinear functions of many variables, smooth interaction surfaces between variables (e.g. latitude&#x2F;longitude), handle repeated measurements and grouping, and deals with many types of outcomes, including continuous, binary yes&#x2F;no, count, ordinal categories, or survival time measurements.<p>All while yielding statistically valid confidence intervals, and typically only taking a few minutes of CPU time even on datasets with hundreds of thousands &#x2F; millions of datapoints.
评论 #39425735 未加载
评论 #39426084 未加载
评论 #39428871 未加载
评论 #39426791 未加载
评论 #39425338 未加载
评论 #39425349 未加载
评论 #39426527 未加载
评论 #39426177 未加载
t_mann超过 1 年前
Statistics is more than hypothesis testing, but you&#x27;ll get surprisingly far without straying too far from linear models - I remember a Stats prof saying &#x27;most of classical Statistics is GLM [0]&#x27;<p>[0] <a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Generalized_linear_model" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Generalized_linear_model</a>
评论 #39423203 未加载
评论 #39422618 未加载
评论 #39424076 未加载
simulo超过 1 年前
I knew the linked-in-the-article <a href="https:&#x2F;&#x2F;lindeloev.github.io&#x2F;tests-as-linear&#x2F;" rel="nofollow">https:&#x2F;&#x2F;lindeloev.github.io&#x2F;tests-as-linear&#x2F;</a> which is also great. A bit meta on the widespread use of linear models: &quot;Transcending General Linear Reality&quot; by Andrew Abbott, DOI:10.2307&#x2F;202114
评论 #39421870 未加载
评论 #39421849 未加载
richrichie超过 1 年前
You can state and prove theorems with linear models. You can do inference and testing. This means papers and academics naturally love them. And therefore, they are everywhere.<p>Not the case with non-linear models. We need to throw computers at them.
评论 #39425299 未加载
jwilber超过 1 年前
Another fun stats X is really Y:<p>estimating the Area Under the Curve metric (AUC) is equivalent to the Wilcoxon-Mann-Whitney test!<p><a href="https:&#x2F;&#x2F;rmets.onlinelibrary.wiley.com&#x2F;doi&#x2F;abs&#x2F;10.1256&#x2F;003590002320603584" rel="nofollow">https:&#x2F;&#x2F;rmets.onlinelibrary.wiley.com&#x2F;doi&#x2F;abs&#x2F;10.1256&#x2F;003590...</a>
SubiculumCode超过 1 年前
I find it irritating that the article mentions repeated measures, but does not try them, much less a mixed effects model. Yes, they are linear models with more parameters, but doing it in lm would be a special kind of madness
matteoraso超过 1 年前
A cool thing about linear models is that they can be used to model non-linear correlations by using transformations. For example, an exponential function can be made linear if you just take the logarithm.
评论 #39424723 未加载
评论 #39424758 未加载
btdmaster超过 1 年前
I thought nonlinearity was very important to be able to make a larger model better than a smaller one? Like so important that tom7 made a half-joke demo with it: <a href="https:&#x2F;&#x2F;yewtu.be&#x2F;watch?v=Ae9EKCyI1xU" rel="nofollow">https:&#x2F;&#x2F;yewtu.be&#x2F;watch?v=Ae9EKCyI1xU</a>
评论 #39422116 未加载
评论 #39422548 未加载
评论 #39421689 未加载
kqr超过 1 年前
This is why I feel like I couldn&#x27;t become a credible AI&#x2F;ML consultant. I would just throw everything into a linear model, make some progress, and call it a day.
评论 #39427106 未加载
cs702超过 1 年前
Maybe it&#x27;s because to scientists and statisticians comfortable with linear algebra, everything looks like a linear model.
dboreham超过 1 年前
Hmm...I thought everything was an Eigenfunction.
评论 #39422143 未加载
chmaynard超过 1 年前
RSS feed: <a href="https:&#x2F;&#x2F;danielroelfs.com&#x2F;blog&#x2F;index.xml" rel="nofollow">https:&#x2F;&#x2F;danielroelfs.com&#x2F;blog&#x2F;index.xml</a>
funcDropShadow超过 1 年前
Everything is linear if plotted log-log with a fat magic marker (Mar&#x27;s Law) [1].<p>[1]: <a href="https:&#x2F;&#x2F;spacecraft.ssl.umd.edu&#x2F;akins_laws.html" rel="nofollow">https:&#x2F;&#x2F;spacecraft.ssl.umd.edu&#x2F;akins_laws.html</a>
ofrzeta超过 1 年前
I thought everything was a power function.
评论 #39423293 未加载
评论 #39423229 未加载
sva_超过 1 年前
(2022)
2-718-281-828超过 1 年前
isn&#x27;t every non linear model fundamentally linear if it is run on a von neumann computer?