Simple Poisson model about stan HOT 2 CLOSED

The problem is that Stan, unlike BUGS and R,
distinguishes between integer and floating-point (real)
types.

Specifically, your declaration

real Y[N];

is incompatible in Stan with its use:

Y[i] ~ poisson(a + b*X[i]);

The variate in a Poisson is required to be an
integer type.

You should be able to fix the problem by declaring:

int Y[N];

If your Y aren't integers, you can't technically use a Poisson
probability mass function.

You could define a probability density function that's Poisson-like
using lp__ as described in the manual section on writing your
own probability functions in Stan.

• Bob

On 2/12/13 5:43 PM, rszoh wrote:

All,

I have the following code that fits a Poisson model to a data:

library(rstan)
Bcca_mod <- '
data {
int N; // sample size
real Y[N];
real X[N];
}
parameters {
real a;
real b;
}
transformed parameters {

}
model {
a ~ gamma(1,1);
b~ gamma(1,1);
for(i in 1:N)
Y[i] ~ poisson(a + b*X[i]);
}
'
N=20

X = seq(0,20,length.out=N)/20
a=1.5
b=3
Y = apply(matrix(a+b*X,ncol=1),1,rpois,n=1)

bcca_dat<-list("Y"=Y,"N"=N,"X"=X)
fit <- stan(model_code = Bcca_mod, data = bcca_dat, iter = 1000, chains = 2)

Note that I have the right constraints on a and b and Y even though they don't show.
When I run the above code I get this error message:

Error in stanc(file = file, model_code = model_code, model_name = model_name, :
failed to parse Stan model 'Bcca_mod' with error message:
EXPECTATION FAILURE LOCATION: file=input; line=18, column=1

Y[i] ~ poisson(a + b*X[i]);
^-- here

DIAGNOSTIC(S) FROM PARSER:
no matches for function name="poisson_log"
arg 0 type=real
arg 1 type=real
unknown distribution=poisson
Parser expecting:

Does it mean that stan does not yet support the poisson distribution??

Any help will be appreciated.
Thanks

—
Reply to this email directly or view it on GitHub #13.

from stan.

I realized after another close look a the code that Y was declared real instead of Int. Now things work just perfect!
Many thanks!

from stan.