K-fold predictive validation of survival (Matlab Interface)

[avehtari]

@Ashwani_Jha

Shouldn’t se(elpd_loo)=1/n(sqrt(…)) ? Otherwise se will simply get bigger se for larger datasets?

Remember that elpd_loo was defined as sum of elpd_loo,i’s and thus se of that sum should get bigger.

@Ashwani_Jha

but couldn’t work out how to apply it to the survival model which has two X’s (censored/ uncensored) parameterised by the same betas. Would it be correct to perform QR on each X individually and model two beta_tildes (censored/ uncensored).

No Do the QR transformation for X which has covariates for both, and then divide the transformed covariates to two parts.

@Ashwani_Jha

plus 100 warmup

That seems quite short and may affect the the performance of the adaptation of NUTS algorithm.