Add variable importance for survival models (survivalmodels package)

[ilagith]

Thank you a lot for this beautiful package!

It would be great to compute variable importance for models implemented through survivalmodels package, in partiucular for neural networks such as deepsurv.

Thanks!

[bgreenwell]

Hi @ilagith, thanks for the note, and glad you find the package useful. In theory you should be able to use one of the model-agnostic approaches (e.g., method="permute"). The challenge is in defining the prediction wrapper and corresponding metric function. For example, in ordinary regression, you might define the metric as RMSE. For survival models, I think the Brier score is an appropriate metric in some cases. If you have a simple example with code, I can try to put together a solution. WOuld be nice for the documentation as well or a vignette.

[bgreenwell]

@ilagith feel free to reopen if you share an example and still would like this functionality.