This deprecates arguments lambda_min_ratio, nlambda, and thresh of varsel() and cv_varsel(). Instead, varsel() and cv_varsel() gain a new argument called search_control which accepts any control arguments for the search (as a list). Thus, former arguments lambda_min_ratio, nlambda, and thresh should now be specified via search_control (but note that search_control is more general because it also accepts control arguments for a forward search).
The main reason for this change is to allow different tuning parameters (= control arguments) for the <refmodel_object>$div_minimizer() function in search_forward() and perf_eval(). However, this change is also the most straightforward solution to ensure that fold-wise searches from cv_varsel.vsel() use the same tuning parameters as the previously run full-data search (output element args_search of vsel objects did not take ... into account).
Note that search_control = list() can be specified when the defaults of the underlying draw-wise divergence minimizer(s) should be used in a (forward) search, but not in the performance evaluation.
This deprecates arguments
lambda_min_ratio,nlambda, andthreshofvarsel()andcv_varsel(). Instead,varsel()andcv_varsel()gain a new argument calledsearch_controlwhich accepts any control arguments for the search (as alist). Thus, former argumentslambda_min_ratio,nlambda, andthreshshould now be specified viasearch_control(but note thatsearch_controlis more general because it also accepts control arguments for a forward search).The main reason for this change is to allow different tuning parameters (= control arguments) for the
<refmodel_object>$div_minimizer()function insearch_forward()andperf_eval(). However, this change is also the most straightforward solution to ensure that fold-wise searches fromcv_varsel.vsel()use the same tuning parameters as the previously run full-data search (output elementargs_searchofvselobjects did not take...into account).Note that
search_control = list()can be specified when the defaults of the underlying draw-wise divergence minimizer(s) should be used in a (forward) search, but not in the performance evaluation.