Sparse regression is the problem of estimating a quantity of interest using a linear model that selects only a small subset of the available features. We show that l1-norm regularization, also known as the lasso, offers an effective solution to this problem. By analyzing a simulated and a real-world dataset, we explore the lasso’s behavior and study why it is well-suited for promoting sparsity.
Probability and statistics book: https://a.co/d/7k259eb
Website with free preprint, videos, slides and solutions to exercises: https://www.ps4ds.net
Slides for this video: https://github.com/cfgranda/ps4ds/blob/main/slides/regression%20and%20classification/sparse_regression_handout.pdf
Code: https://github.com/cfgranda/ps4ds/blob/main/regression_classification/two_features_lasso.ipynb
https://github.com/cfgranda/ps4ds/blob/main/regression_classification/two_features_lasso_cost_function.ipynb
https://github.com/cfgranda/ps4ds/blob/main/regression_classification/temperature_linear_regression.ipynb
