Horseshoe prior error in rstan

Modeling
1

Can you please fix why my code produces an error??

data {
  int<lower=1> n; // Number of data
  int<lower=1> p; // Number of covariates
  matrix[n,p] X;  // n-by-p design matrix
  real y[n];      // n-dimensional response vector
}

parameters {
  vector[p] beta;
  vector<lower=0>[p] lambda_sq;
  vector<lower=0>[p] omega;
  real<lower=0> tau_sq;
  real<lower=0> eta;
  real<lower=0> sigma;
}

transformed parameters {
  vector[n] theta ;
  theta = X * beta;
}

model {
  // Origianal horseshoe by carvalho + Parameter expansion on half-Cauchy
  //lambda ~ cauchy(0, 1) is equivalent with;
  omega ~ inv_gamma(1 / 2, 1);
  lambda_sq ~ inv_gamma(1 / 2, 1 ./ omega);
  //tau ~ cauchy(0, 1) is equivalent with;;
  eta ~ inv_gamma(1 / 2, 1);
  tau_sq ~ inv_gamma(1 / 2, 1 / eta);
  beta ~ normal(0, sigma * sqrt(tau_sq) * sqrt(lambda_sq) ); 
  y ~ normal(theta, sigma);
}
2

It’s easier if you post what the error message is.

3

Do you know how to make 1/2 to be interpreted as 0.5?

It seems that error comes from “integer division implicitly rounds to integer. Found int division: 1 / 2”

4

You can either do 0.5 or 1/2.0, but these are warnings, not errors.

5

If I use 0.5 or 1/2.0, then R aborted.

6

Can you share your R code and maybe some test data? It helps also if you can post the result of sessionInfo().

7

horseshoe_pe.stan (773 Bytes)

8

Please check this: horseshoe_pe.stan (773 Bytes)

You may use any (ordinary) linear regression for the y and X.