Sampling from a matrix in STAN

Modeling
, ,
22

You can’t define functions within functions, you need to declare the rng function separately and then call it within your function

1 Like
23

Hi @andrjohns @bbbales2 @jsocolar ,

I got the following error when running the model.

Found negative dimension size in variable declaration; variable=Gender; dimension size expression=n; expression value=-2147483648

This is the stan code defined in a user-defined _rng function.

vector[n] Gender;
for (i in 1:n){
      Gender[i] = binomial_rng(1,0.5);
}

Can someone explain me why this error happens

Thank you

24

Presumably because n is negative?

25

Hi @jsocolar ,

I don’t still know how to solve it because I put n=100

Thanks

26

Can you post your full stan code? This error indicates that n has not been given a value before it’s been used

27

Hi @andrjohns,

That means do I need to give an initial value for n? Also, I have a new error

algebra_solver: parameter vector[1] is nan, but must be finite

Okay, please see attached stan code and help me to make it run.
model2.stan (8.9 KB)

Thank you

28

The error for n is caused by this section:

matrix data_rng(real m, real mu_diag, real mu_reg, real mu_dist){
    int n ;
  ...
    vector[n] Gender;

You declared the int variable n and are using it for indexing, but you haven’t given it a value, so Stan doesn’t know what size these vectors should be or what length to loop over

29

Thank you Andrew.
Okay, I set it n = 100 within my user-defined function data_rng (is it correct?) and run the model.
Then it gave me this error algebra_solver: parameter vector[1] is nan, but must be finite

This is the first time I am using an algebraic solver in stan and I don’t much about it. So, can you please explain to me how to solve it?

Thank you

30

That error indicates that at least one of the values being passed to the algebra_solver function is nan, try using the print function to print the values of the parameters that you’re using as inputs, to see which parameters are nan

1 Like