Why L1 Regularization Produces Sparse Solutions

EasyClassical ML

Reported in interviews at Amazon Citadel Inmobi Proxify Renaissance Technologies Schonfeld Strategic Advisors

Problem statement

Why can lasso set regression coefficients exactly to zero while ridge usually only shrinks them toward zero? Explain the constraint geometry and the optimality conditions at a zero coefficient. Use a two-dimensional example.

Show hint

What does the feasible region of each penalty look like when you draw it in two dimensions?

Relevant roles: ML Engineer, Applied Scientist, Research Scientist and Data Scientist.

Related free problems