Lasso and Ridge regression models in R: Introduction
summary
Hands-on R demo using the WEF dataset, glmnet and cv.glmnet to show that regularization reduces overfitting: ordinary lm has adjusted R² = 0.182, Ridge shrinks coefficients (goods market efficiency 0.600 → 0.568, market size 0.134 → 0.131) and Lasso can drive some coefficients to zero. Cross-validation picks lambda (cv.glmnet gives lambda.min ≈ 0.066 for Ridge, ≈ 0.0022 for Lasso) and Elastic Net mixes them via Alpha between 0 (Ridge) and 1 (Lasso). Code-level steps and plots for coefficient paths and cv.glmnet finds lambda via cross-validation are provided.
AI-written summary · about
chapters (9)
- 0:00Lasso and Ridge in R
- 1:03Theory of Lasso and Ridge
- 24:29Linear regression limits
- 36:16Overfitting and regularization
- 42:56Ridge regression and Lambda
- 1:08:32Ridge regression demo in R
- 1:32:57Interpreting regression results
- 1:51:37Lasso regression intro
- 2:03:52Elastic net and model comparison
related talks
25mAki720
32mRidge Regression and Deterministic Ridge Leverage Score Sampling
31mLogistic Regression Behind the Scenes
46mLearning Functions and Sets with Spectral Regularization
45mPyglmnet: A Python package for elastic net regularized generalized linear model
49mSo you think you know about linear regression ...
44mActive Regression via Linear-Sample Sparsification
01h 23mImplicit Regularization II