Unexpected Results when using WAIC to Compute Log Predictive Accuracy

I can run loo on the test data set without problem - are you on master?
Two quick comments:

  1. When I run the code above I see a lot of warning for both WAIC and LOO, this is an indication of problem in the computation and you cannot trust the result.
  2. I would suggest you to set the number of predictors to a lower dimension - it is likely that your model is overfitting with few data points.
1 Like