random warning on unit test

[tjmahr]

Sometimes I get the following warning when running tests

test-ppc-test-statistics.R:30: warning: ppc_stat returns ggplot object
Computation failed in `stat_bin()`:
`binwidth` must be positive

The specific test is

  expect_gg(ppc_stat(y2, yrep2, stat = "prop0"))

But this doesn't happen every time. We need to bin down this nondeterminism.

[jgabry]

I haven't seen that before, but my first guess is that this is because the seed for generating the data is set on line 9 instead of line 1:

bayesplot/tests/testthat/data-for-ppc-tests.R

Lines 5 to 12 in b27d9b7

	y2 <- rpois(30, 1)
	yrep2 <- matrix(rpois(30, 1), ncol = 30)
	group2 <- rep(1, 30)
	set.seed(11172017)
	vdiff_y <- rnorm(100)
	vdiff_yrep <- matrix(rnorm(2500), ncol = 100)
	vdiff_group <- gl(4, 25, labels = LETTERS[1:4])

(Same issue with the script generating the data for the MCMC plot tests. )

I can't be sure this is the problem because I haven't seen that error, but either way we should probably be setting the seed before simulating data for the tests. I thought we were, but I guess not. Sorry about that.

[jgabry]

Alternatively, we could save the data for these tests in an rds or rda file and just load it in.

[jgabry]

Ok so this is related to tidyverse/ggplot2#3043 and is fixed by tidyverse/ggplot2#3047. In d5bd1a8 I've also added a seed when we generate data for the tests.

I think it's safe to close this now.