While trying to create a test case for non-linearity handling, we accidentally created a case to demonstrate instabilities in high-dimensional linear models. The fact it happened easily and comparison to real data in terms of spectra suggests this is a generic problem that needs to be handled by any candidate model, and thus a valuable test case to have in itself.
[x] Factor out linear and kmeans in models (negative control)
[x] Refactor current accidental case into a lin reg case
[x] add a positive control (e.g. cv-based ridge or pca)
While trying to create a test case for non-linearity handling, we accidentally created a case to demonstrate instabilities in high-dimensional linear models. The fact it happened easily and comparison to real data in terms of spectra suggests this is a generic problem that needs to be handled by any candidate model, and thus a valuable test case to have in itself.
models
(negative control)