gqs() error with length 1 vector or array parameters

Summary:

gqs() gives an error when one of the parameters is a vector or array of length 1

#> Exception: Variable d missing  (in 'model17545a0741ce_43effcd60c5e7661f052cf84d7a670d3' at line 5)

Description:

See above.

Reproducible Steps:

The following code works when nd = 2 or greater, but fails when nd = 1.

sm <- stan_model(model_code =
"data {
  int<lower=1> nd;
}
parameters {
  vector[nd] d;
}
generated quantities {
  real d_new[nd] = normal_rng(d, 1);
}")

# Data
nd <- 1 # works with 2+, fails with 1

# Dummy draws
iter <- 1000
d <- matrix(rnorm(iter * nd), ncol = nd, nrow = iter)
colnames(d) <- paste0("d[", 1:nd, "]")

# Generated quantities
gqs(sm, draws = d, data = list(nd = nd))

Current Output:

When nd = 1, the gqs() call fails with the following output:

> gqs(sm, draws = d, data = list(nd = nd))
#> Exception: Variable d missing  (in 'model17545a0741ce_43effcd60c5e7661f052cf84d7a670d3' at line 5)
#> 
#> Inference for Stan model: 43effcd60c5e7661f052cf84d7a670d3.
#> 1 chains, each with iter=1000; warmup=0; thin=1; 
#> post-warmup draws per chain=1000, total post-warmup draws=1000.
#> 
#>          mean se_mean sd 2.5% 25% 50% 75% 97.5% n_eff Rhat
#> d_new[1]    0     NaN  0    0   0   0   0     0   NaN  NaN
#> 
#> Samples were drawn using  at Wed Jan 12 08:58:06 2022.
#> For each parameter, n_eff is a crude measure of effective sample size,
#> and Rhat is the potential scale reduction factor on split chains (at 
#> convergence, Rhat=1).

The error occurs whether vector[nd] d or array d[nd] are specified in the parameters block.

Inspecting the gqs() function, the draws matrix seems to be constructed correctly and passed to sampler$standalone_gqs - so perhaps this is a core stan issue rather than rstan?

Expected Output:

When nd = 2 or greater, the gqs() call outputs as expected:

> gqs(sm, draws = d, data = list(nd = nd))
#> Inference for Stan model: 43effcd60c5e7661f052cf84d7a670d3.
#> 1 chains, each with iter=1000; warmup=0; thin=1; 
#> post-warmup draws per chain=1000, total post-warmup draws=1000.
#> 
#>           mean se_mean   sd  2.5%   25%   50%  75% 97.5% n_eff Rhat
#> d_new[1]  0.00    0.04 1.41 -2.73 -0.92  0.05 0.94  2.81   988    1
#> d_new[2] -0.02    0.04 1.33 -2.77 -0.85 -0.02 0.83  2.56   936    1
#> 
#> Samples were drawn using  at Wed Jan 12 08:58:15 2022.
#> For each parameter, n_eff is a crude measure of effective sample size,
#> and Rhat is the potential scale reduction factor on split chains (at 
#> convergence, Rhat=1).

RStan Version:

The version of RStan you are running (e.g., from packageVersion("rstan"))

> packageVersion("rstan")
#> [1] ‘2.21.2’

R Version:

The version of R you are running (e.g., from R.version.string)

> R.version.string
#> [1] "R version 4.1.2 (2021-11-01)"

Operating System:

Windows 10

Start with the gqs() entry point and the sampler$standalone_gqs path mentioned in the issue, then reproduce the example with nd = 1 and nd = 2. Trace how the single-column draws matrix is passed to generated quantities and verify that both vector[nd] and array declarations complete without the missing-variable error.