79279849

Date: 2024-12-14 00:19:14
Score: 2
Natty:
Report link

Your L1-regularized logistic regression (a.k.a. Lasso penalty) might pick different subsets of correlated features across runs because L1-regularization enforces sparsity in a somewhat arbitrary way when correlation is present. Zeroed-out coefficients aren’t necessarily “worthless”; they may just be overshadowed by a correlated feature that the model latched onto first.

Reasons:
  • No code block (0.5):
  • Single line (0.5):
  • Low reputation (1):
Posted by: Hui Tang