coefficients from each run

[richabatra]

Is it possible to fetch the coefficients of the variables from each run? It is great to know the stable associations and would be even better to know the direction and degree of relationship between response and variables.

[hofnerb]

Not directly, no.

But you could write a modified fitfun (see https://github.com/hofnerb/stabs/blob/master/R/fitfuns.R) that returns the coefficients in addition to selected and path (or if that doesn't work, I didn't think deep enough about that issue, instead of one of the slots). Returning it additionally might need a few changes in the stabsel (and/or run_stabsel, see https://github.com/hofnerb/stabs/blob/master/R/stabsel.R#L273) function to allow forwarding/aggregating the relevant slot. You could provide a patch for this!

Alternatively, you can write your own resampling function to obtain the relevant output and simply use the function stabsel_parameters to obtain the relevant parameter constalation for q, cutoff and PFER for the given data set and additional assumptions.

[richabatra]

alright, I try. Thanks for the input.

[richabatra]

your pointers were great! For starters, I extended it for glmnet.lasso, such that it can return a matrix, n (=features) by p (=mean, sd, range of the coefficients) for each lambda.
would you say this is a valid approach? do you have other suggestions, I am happy to learn.