A model with a random effect fails at parameter initialization in 2.1.18. It reproduces on the counts_obj, so it doesn't seem data-specific:
library(sccomp); library(dplyr)
data("counts_obj")
counts_obj |>
mutate(donor = sample) |>
sccomp_estimate(~ type + (1 | donor), .sample = sample,
.cell_group = cell_group, .abundance = count,
inference_method = "hmc", cores = 1)
Error:
Chain 1 Unrecoverable error evaluating the log probability at the initial value.
Chain 1 Exception: mismatch in dimension declared and found in context; processing stage=parameter initialization; variable name=random_effect_raw; position=1; dims declared=(20,36); dims found=(20,35) (in '/tmp/RtmpBn8kxm/model-8cff8636a24e8.stan', line 348, column 2 to column 93)
Error: Fitting failed. Unable to retrieve the metadata.
Without random effect it works without any issues.
For now I will go back to 1.10.0 where it works. Any advice is appreciated.
Environment:
-sccomp 2.1.18
-cmdstanr 0.9.0, CmdStan 2.39.0
-R 4.4.3
A model with a random effect fails at parameter initialization in 2.1.18. It reproduces on the
counts_obj, so it doesn't seem data-specific:Error:
Without random effect it works without any issues.
For now I will go back to 1.10.0 where it works. Any advice is appreciated.
Environment:
-sccomp 2.1.18
-cmdstanr 0.9.0, CmdStan 2.39.0
-R 4.4.3