Noting which tests fail in feature/survival branch (mostly frailtymod)

Summary:

There are a dozen of failing tests in the feature/survival branch, specific to the test_stan_surv.R function. All appear to be related to the frailty model.

These are run against a WIP branch at https://github.com/jburos/rstanarm/tree/review-failing-tests, which is a fork of feature/survival branch in this repo that:

1. removes a few of the JM or mvmer-specific changes in the feature/survival branch, and
2. updates the syntax for tests to be consistent with current conventions.

Note: Some of these may be addressed by the sparse matrix fixes -- analogous to https://github.com/stan-dev/rstanarm/commit/22d60534e52ab50f7e521d8a77aae0d6aa8114c2 but for the survival model -- which I have in a separate branch. Reporting this here for the sake of transparency.

Description:

These are the failures:

── Failure (test_stan_surv.R:776:3): (code run outside of `test_that()`) ───────
get_ests(modfrail)[[i]] not equal to true[[i]].
1/1 mismatches
[1] 0.521 - -2.3 == 2.82

── Failure (test_stan_surv.R:776:3): (code run outside of `test_that()`) ───────
get_ests(modfrail)[[i]] not equal to true[[i]].
1/1 mismatches
[1] 0.487 - 0.3 == 0.187

── Failure (test_stan_surv.R:776:3): (code run outside of `test_that()`) ───────
get_ests(modfrail)[[i]] not equal to true[[i]].
1/1 mismatches
[1] -1.01 - -2.3 == 1.29

── Failure (test_stan_surv.R:776:3): (code run outside of `test_that()`) ───────
get_ests(modfrail)[[i]] not equal to true[[i]].
1/1 mismatches
[1] 0.765 - 1 == -0.235

── Failure (test_stan_surv.R:776:3): (code run outside of `test_that()`) ───────
get_ests(modfrail)[[i]] not equal to true[[i]].
1/1 mismatches
[1] 0.486 - 0.3 == 0.186

── Failure (test_stan_surv.R:776:3): (code run outside of `test_that()`) ───────
get_ests(modfrail)[[i]] not equal to true[[i]].
1/1 mismatches
[1] 0.698 - -2.3 == 3

── Failure (test_stan_surv.R:776:3): (code run outside of `test_that()`) ───────
get_ests(modfrail)[[i]] not equal to true[[i]].
1/1 mismatches
[1] 0.797 - 1 == -0.203

── Failure (test_stan_surv.R:776:3): (code run outside of `test_that()`) ───────
get_ests(modfrail)[[i]] not equal to true[[i]].
1/1 mismatches
[1] 1.25 - -2.3 == 3.55

[ FAIL 12 | WARN 649 | SKIP 0 | PASS 381 ]

Reproducible Steps:

After cloning the fork mentioned above:

git clone https://github.com/jburos/rstanarm/tree/review-failing-tests
cd rstanarm
git checkout review-failing-tests

Run the survival-model related tests.

> devtools::load_all()
> testthat::test_file('tests/testthat/test_stan_surv.R')

RStanARM Version:

> packageVersion("rstanarm")
[1] ‘2.21.2’

R Version:

> getRversion()
[1] ‘4.2.0’

Operating System:

─ Session info ───────────────────────────────────────────────────────────────
 setting  value
 version  R version 4.2.0 (2022-04-22)
 os       macOS 13.1
 system   x86_64, darwin17.0
 ui       RStudio
 language (EN)
 collate  en_US.UTF-8
 ctype    en_US.UTF-8
 tz       America/New_York
 date     2023-02-07
 rstudio  2022.12.0+353 Elsbeth Geranium (desktop)
 pandoc   2.19.2 @ /Applications/RStudio.app/Contents/Resources/app/quarto/bin/tools/ (via rmarkdown)

Start with the feature/survival branch and run tests/testthat/test_stan_surv.R after loading the package with devtools::load_all(). Inspect the failures around line 776 and the modfrail estimates, comparing them with the sparse-matrix changes noted in the issue. Done means the twelve frailty-related failures are explained and the survival test file passes or has justified expectations.