WAIC calculations

[florianhartig]

add to the WAIC function the option to read in a pointwise logLikelihood matrix from another program, e.g. Jags.

While at it

• check that calculations are identical to WAIC in loo package
• consider adding WBIC
• the line with i <- sample.int(nrow(x),numSamples,replace=TRUE) # should replace really be true? is weird, why don't we directly draw numSamples in getSample? ...also, I don't really know why the replace = T option was chosen. Probably it doesn't make a difference, but I guess the normal getSample should be fine (which has replace = F)

[jds485]

Hi Florian et al.,

I was just checking the WAIC output from BayesianTools and loo packages. In line 62 of the WAIC function, can you confirm the expected dimensions of the pointwise log-likelihood matrix? I found that I needed to remove the matrix transpose in this line for the results to essentially match using BayesianTools WAIC2 output and the loo waic function.

Thanks! Jared

[florianhartig]

Hi Jared, thanks for that! I had planned to do some long overdue maintenance on this package in the next days anyway, and will bump this to the top of the list!