Virgin Non-Parametric Regression vs Chad Linear Regression Comparison
Description
A classic 'Virgin vs Chad' meme format comparing non-parametric regression (left, 'The virgin') with linear regression (right, 'THE CHAD'). The virgin non-parametric regression is depicted with a scattered, messy data cloud and numerous criticisms: 'Requires a tonne of observations for convergence', 'Heavily depends on hyper-parameters', 'Does not have a closed-form solution', 'Terribly inefficient high-level implementations slower than Crysis on Pentium III', 'Dies a horrible death with 10 regressors from the curse of dimensionality'. The Chad linear regression shows clean, colorful clustered data points with praise: 'With 20 data points ends up being published in top macroeconomic journals', 'Works in one step without fine-tuning', 'Can be computed analytically with paper and pencil', 'Optimised to the Moon and beyond in linear algebra libraries', 'Inference can be done with one subtraction and one division by a toddler'
Comments
18Comment deleted
Linear regression: the only algorithm that gets published in top journals, runs on a toddler's arithmetic skills, still outperforms your 47-layer neural network on tabular data, and has the audacity to be interpretable while doing it
Your model has so many hyperparameters it needs a YAML file and a Kubernetes cluster to run. My model is a straight line, and I derived it on a napkin during lunch
Why burn GPU hours to torture a kernel smoother when good old OLS can ship to prod before your coffee cools?
The real Chad move is using linear regression on clearly non-linear data and just adding 'robust standard errors' to the paper title - it's been getting economists tenure since 1970
Linear regression: the only model where violating every assumption still gets you published in top journals, while non-parametric methods need a supercomputer and a PhD thesis just to fit 10 variables
Non-parametrics for arXiv glory; linear regression for the boardroom demo that actually ships
Nonparametrics: a grid search to estimate one slope. OLS: add “robust” and the reviewer stops asking questions
When I was in university, we have robot with 2d lidar. I wrote program that use linear regression to extract line info from lidar data and perform navigation using this lines instead of point cloud. that was extremely effective especially in our environment, where our track was build from paper boxes. Comment deleted
can u tell me how did u wrote the linear regression? do u have github repo? Comment deleted
dude, this is school level knowledge Comment deleted
oh, ok Comment deleted
but i need it for my onnx model Comment deleted
just Google the algorithm. it's very basic Comment deleted
ok Comment deleted
I am sure they have builtin linear regression, lol Comment deleted
lol Comment deleted
so ok Comment deleted
Wtf am i looking at Comment deleted