Exception: mismatch in number dimensions declared and found in context

General
1

I am trying to use a self-defined distribution in Stan to get the parameter posterior distributions. Everything works out well except when the vector has the length 1, Stan will not work. I found some similar topics, but I don’t really understand the reason why Stan does not work in this condition. Here are my Stan code and some sample data:

plpstan11 = '
functions{
  real nhpp_log(vector t, real beta, real theta, real tau){
    real loglikelihood;
      vector[num_elements(t)] loglik_part;
      for (i in 1:num_elements(t)){
        loglik_part[i] = log(beta) - beta*log(theta) + (beta - 1)*log(t[i]);
      }
      loglikelihood = sum(loglik_part) - (tau/theta)^beta;
    return loglikelihood;
  }
}
data {
  int<lower=0> n; //total # of obs
  real<lower=0> tau;//truncated time
  vector<lower=0>[n] t; //failure time
}
parameters{
  real<lower=0> beta;
  real<lower=0> theta;
}
model{
  t ~ nhpp(beta, theta, tau);
//PRIORS
  beta ~ gamma(1, 1);
  theta ~ gamma(1, 0.01);
}
'

library(rstan)
stan_yes = list(n = 6, tau = 657, t = c(251, 397, 406, 451, 465, 593))
stan_no = list(n = 1, tau = 657, t = 253))

fitplp <- stan(
  model_code=plpstan11, model_name="NHPP", data=stan_no, 
  iter=1000,warmup = 500, chains=1, seed = 123
)

The stan code will work out well if the length of t is more than 1 (using the data stan_yes). However, Stan will report the following error message only if t is a vector with length 1 (using the data stan_no):

Error in new_CppObject_xp(fields$.module, fields$.pointer, …) :
Exception: mismatch in number dimensions declared and found in context; processing stage=data initialization; variable name=t; dims declared=(1); dims found=() (in ‘model352055946207_NHPP’ at line 16)
failed to create the sampler; sampling not done

I found two similar posts here:

But from my limited understanding, none of them clearly explain the mechanism behind this error. Could anyone explain this error? Thank you!

1 Like
2

You probably just need to do as.array(t) when you construct the list of data in R when t is of length 1.

2 Likes
3

Thanks Ben! That works. I’ve seen similar answers in the two links I posted.
Is there any reason why as.array(t) works?

If I inspect into the data list, there seems to no difference before and after as.array()

4

Stan will interpret t as a scalar if you do not do as.array, leading to the error message you saw.

5

Got it. Thanks Ben!

6

why only “t” since both “n” and “tau” have length 1?

7

Those are declared as vectors that happen to have a size of 1, rather than scalars.