Brms kfold by response after fitting full kfold

I have run a kfold cross validation on a multi-response brms model as follows:

kf_mod1nosp <- kfold(model1_nosp, save_fits = TRUE, K=10)

Given that I have 3 response variables, is it possible to extract a kfold ic for each response variable given that I have the fits saved in the kfold object? Or do I have to rerun kfold with "resp = " specified for each response separately?

  • Operating System: Windows 10 x64
  • brms Version: version 2.10.0

Hi, I might be completely wrong here but can’t you use resp = c("var1", "var2")?

I assume I’d have to rerun the kfold analysis then (which took a month). I’m wondering if there is any way to extract the single response information from a full kfhold analysis.

Since kfold uses the elpd measure for model fit, which in turn uses the joint likelihood of all response variables per observation, I am afraid there is no way to compute the value for a subset of responses post-hoc.