5 Testing

All tests go in tests/testthat. Test everything you could possibly think of. If you think it should be a test, then that probably means it should be a test. Arguably, before writing any code you should write (at the very least) some basic tests first.

Make sure you add a test file for the model you’re including as tests/testthat/test_stan_*.R. Also don’t forget to add the relevant tests for the methods associated with your model in the other test files.

For speed, most of the tests should be specified using algorithm = 'optimizing', QR = TRUE, and around 200 iterations.

You should have tests for the following,

• All variations of model specification and comparison with the package you’re emulating. (e.g. model with a constant, without a constant, etc.)
• Test predict/posterior_predict with and without new data.
• Test that loo and compare work on the stanreg object.

Run a comprehensive test of rstanarm often. Especially if you’re altering Stan files. This will help you catch any bugs early on (which means they’ll be easier to fix).

This script should be sufficient (at the time of writing) to run all the tests (excluding the vignettes):

### script to run all rstanarm tests locally

library(rstanarm)
# library(rstantools)
# library(bayesplot)

remove(list=ls())
### run prerequisite functions
#
example_model <-
  stan_glmer(cbind(incidence, size - incidence) ~ size
  + period + (1|herd),
             data = lme4::cbpp, family = binomial,
             # this next line is only to keep the example small in size!
             chains = 2, cores = 1, seed = 12345, iter = 500)
#
last_dimnames <- function(x) {
  ndim <- length(dim(x))
  dimnames(x)[[ndim]]
}

### run tests
devtools::test()