Support all pp_check types for censored y #1327
Conversation
tests/brmsfit_examples.R
Outdated
| @@ -84,16 +84,24 @@ brmsfit_example6 <- brm( | |||
| stan_model_args = stan_model_args, rename = FALSE | |||
| ) | |||
|
|
|||
| brmsfit_example7 <- brm( | |||
There was a problem hiding this comment.
These internal models take quite some space so I would like to avoid saving another one. Can you add this test to the local tests in tests/local/? Perhaps there is already a censored model.
There was a problem hiding this comment.
Good idea thanks, have reverted those changes and moved. Though when running the local tests beforehand I noticed the add_rstan_model function fails for that kidney survival example. Is this the same on your computer, and if so, is it okay to ignore for this PR and handle separately?
Test fail:
-- Error (tests.models_new.R:89:3): Survival model from brm doc works correctly ---------------------------
Error in `stanc(file = file, model_code = model_code, model_name = model_name,
verbose = verbose, obfuscate_model_name = obfuscate_model_name,
allow_undefined = allow_undefined, isystem = isystem)`: failed to parse Stan model '530b15df4662e9b168e0f021510d8a97' due to the above error.
Backtrace:
1. brms::add_rstan_model(fit3)
at tests.models_new.R:89:2
4. rstan::stan(...)
at brms/R/brmsfit-helpers.R:761:4
5. rstan::stan_model(...)
6. rstan::stanc(...)
Checking internally looks like the arrays from the cmdstanr backend:
rstan::stan(
+ model_code = stancode(x, threads = threading()),
+ data = standata(x), chains = 0
+ )
SYNTAX ERROR, MESSAGE(S) FROM PARSER:
error in 'modeldb4faf53a2_530b15df4662e9b168e0f021510d8a97' at line 8, column 2
-------------------------------------------------
6: int<lower=1> N; // total number of observations
7: vector[N] Y; // response variable
8: array[N] int<lower=-1, upper=2> cens; // indicates censoring
^
9: int<lower=1> K; // number of population-level effects
-------------------------------------------------
PARSER EXPECTED: <one of the following:
a variable declaration, beginning with type,
(int, real, vector, row_vector, matrix, unit_vector,
simplex, ordered, positive_ordered,
corr_matrix, cov_matrix,
cholesky_corr, cholesky_cov
or '}' to close variable declarations>
Error in stanc(file = file, model_code = model_code, model_name = model_name, :
failed to parse Stan model '530b15df4662e9b168e0f021510d8a97' due to the above error.
Seems to fix if I add rtsan backend to the recompiling code (
Line 762 in 9410572
model_code = stancode(x, threads = threading(), backend = 'rstan'),
^^^^^^^^^^^^^^^^^^^
But I can submit that in different request? Or you check/change? Whatever's easiest
There was a problem hiding this comment.
This is just because of new array syntax in the latest version of Stan (>= 2.29.0). You can safely ignore that here.
hdrab127
force-pushed
the
issue-pp-check-grouped-cens
branch
from
912072e to
a75cce3
Compare
|
Sorry I wasn't sure the cleanest way to revert the |
Codecov Report
@@ Coverage Diff @@
## master #1327 +/- ##
==========================================
- Coverage 86.31% 83.15% -3.16%
==========================================
Files 42 70 +28
Lines 15511 18958 +3447
==========================================
+ Hits 13388 15765 +2377
- Misses 2123 3193 +1070
Continue to review full report at Codecov.
|
|
Thank you! Looks good to me! Merging now! |
Hopefully addresses #1326.
The diff got a bit confused, it's mostly just moving the
ppc_argassignments, likegroup,lw,x,psisabove the censor check, then subsetting them there if they are used.Also allows censored values for the
type = "km_overlay"plot (https://mc-stan.org/bayesplot/reference/PPC-censoring.html).sysdata.rdajust hasbrms:::brmsfit_example7included for the censored tests, I didn't re-run the other example fits.Re-running some of the examples in the issue:
Created on 2022-03-23 by the reprex package (v2.0.1)