Defining priors for k-dimensional arrays

Modeling
1

I want to define priors for a k-dimensional array, such as


parameters{
    array[dim1, dim2, dim3] real param;
}

model {
  param ~ beta(3, 3); 
}

But stan is complaining with an error, I can only define the priors using a loop, for each parameter. I don’t have an issue with defining the loop.


parameters{
    array[dim1, dim2, dim3] real param;
}

model {
  for(i in 1:dim1){
    for(j in 1:dim2){
        for(k in 1:dim3){
            param[i, j, k] ~ beta(3, 3); 
        }
      }
  }
}

But I am worried that it will make the code much slower since it will need to define dim1dim2dim3 priors individually.

Does anyone know if there is a big performance difference? Or if there is an alternative to making a loop?

Thanks :)

2

You can flatten it with to_array_1d:

to_array_1d( param ) ~ beta(3, 3);

1 Like
3

Ooh wow interesting approach, will that have the same result? Do you think the cost of flattening expensive or the same as loop?

4

You can wrap each in the function profile using cmdstanr to see what the difference is: Profiling Stan programs with CmdStanR • cmdstanr

2 Likes