Josef Perktold
Josef Perktold
We could add more robust norms and helper functions like computing relative efficiency at normal distribution. Menezes et al has a long list of norms see also https://github.com/statsmodels/statsmodels/pull/8801#issuecomment-1510076913 and following...
RLM norms define default parameters, but it's not clear how they are/were chosen. SAS mentions that theirs are at 95% relative efficiency for normal model (except for Hampel and Welch)....
see #8801 and #8808 we don't have a unit test in the consistency checks that rho is scaled so that weights(0) = 0 we can just add something similar to...
I'm running into various measure for the closeness of two covariance or scatter matrices. We should get a function with a collection of those. current usecase: checking how well a...
(This is just an idea, I have not seen it in the literature, but I did not look at the high-dimension robust cov literature) Problem As k_vars increase, efficiency of...
context #9227 adding S- and MM-estimators I thinking whether we should change the starting parameters in M-estimators to a robust estimate, currently for CovM. For monotone norm, the minimum is...
GLM.fit_regularized uses `**kwargs` for some penalization keywords but doesn't show what the defaults are. e.g. it is not visible what L1_wt default is ``` defaults = {"maxiter": 50, "L1_wt": 1,...
initial version of 2 fixes for - overparameterized means of random effects - enforce positive scale of random effects variance using exp instead of hard lower bound see #2382 for...
https://documentation.sas.com/doc/en/statcdc/14.2/statug/statug_rreg_details26.htm#statug.rreg.robustreggrd SAS MCD takes into account that cov in subset could be singular. Standard MCD algorithm assumes full rank. pseudo determinant is product of non-zero, strictly positive eigenvalues (where numeric...
The recent robust merges need another review and maybe some enhancements before release (items what I remember right now) - check api - RLMXxx - CovXxx, - cov function like...