Skip to content

Unexpected selection probabilities using cv_varsel (k-fold forward search) - following body fat tutorial #311

@adietzel

Description

@adietzel

Learning the ropes of projpred currently and ran the body fat example (https://avehtari.github.io/modelselection/bodyfat.html) with cross-validated variable selection using cv_varsel().

Setting method to loo I obtained the following error message:

"Error in fixFormulaObject(terms) :
'rho' must be an environment not pairlist: detected in C-level eval"

Tried k-fold (K=10) after as a forward search but obtained unexpected selection probabilities from cv_varsel().
Most selection probabilities above the diagonal are zero rather than one, as in the tutorial on the website.
Selection probabilities now sum up to one for each sub-model size/column.

This would suggest that variables like "abdomen" and "weight" only matter in small models but not models with more variables but I presume something is not quite right here. Any help would be appreciated.
I have encountered the same issue or pattern when running cv_varsel() on the normally distributed data from the toy example in the vignette (https://cran.r-project.org/web/packages/projpred/vignettes/projpred.html).

Any help would be greatly appreciated.
Best

Andreas

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions