ppc_bars (and similar) throws error with negative integers

[jeffpollock9]

Hi, may I ask why ppc_bars (and other discrete outcome functions, I think) throws an error when y or yrep contain negative values? For example I would like to use this function with the skellam distribution as follows:

library(bayesplot)
library(skellam)

n <- 1000
m <- 500

y <- rskellam(n, 1.5, 2.5)
yrep <- matrix(rskellam(m * n, 1.4, 2.6), m, n)

ppc_bars(y, yrep)

which currently outputs:

> ppc_bars(y, yrep)
Error in ppc_bars(y, yrep) : 
  ppc_bars expects only non-negative integers in 'y'.

If I remove the checks, which are:

if (!all_counts(y)) 
    stop("ppc_bars expects only non-negative integers in 'y'.")
if (!all_counts(yrep)) 
    stop("ppc_bars expects only non-negative integers in 'yrep'.")

Then the plot works as expected:

I appreciate that the majority of the time the outcomes are non-negative, but it is not always the case and I don't think that this function should error here. Thanks!

[jgabry]

Thanks for reporting this. You’re quite right, that’s a bug. We’ll relax those checks in the next release.