Lasso Regularisation Path: Soft Thresholding, Coordinate Descent and Sparse Recovery (2D)
Coefficient paths against the penalty, the current coefficient values, test error and active-set charts with Blender-rendered desk props. The same shared model as the 3D board.
Relax the L1 penalty and watch the lasso coefficients leave zero one by one. Three true signals hide among eight correlated features; change noise, correlation and sample size and follow the path, the active set and the expected test error. A Blender-built data board with KKT, orthonormal closed-form and least-squares checks.
Coefficient paths against the penalty, the current coefficient values, test error and active-set charts with Blender-rendered desk props. The same shared model as the 3D board.
2D · HTML5 Canvas 2D · 60 FPS target · runs fully client-side, no install