Syntax error?

Modeling
1

data {
int<lower=0> N; // number of observations
vector[N] x; // predictor variables
vector[N] y; // response variable
}

parameters {
real mu; // intercept
real<lower=0> sigma; // residual standard deviation
vector[N] f; // function values at observed x
}

model {
f ~ bart_custom_lpdf(0, 0, 0.5, 2, x); // BART prior
y ~ normal(f, sigma); // likelihood
sigma ~ gamma(0.5, 0.5); // gamma prior on sigma
}

generated quantities {
vector[N] y_pred; // predicted values
for (i in 1:N) {
y_pred[i] = normal_rng(f[i], sigma);
}
}

functions {
real bart_custom_lpdf(vector f, real a, real b, real tau, vector x) {
// Define custom probability function for bart_custom
// f: function values at observed x
// a: alpha
// b: beta
// tau: tau
// x: predictor variables
int num_trees = 50; // number of trees
int N = size(x);
matrix[N, num_trees] trees;
vector[num_trees] weights;
vector[N] mu;
real lp = 0;

for (j in 1:num_trees) {
  for (i in 1:N) {
    trees[i, j] = normal_rng(0, 1);
  }
  weights[j] = normal_rng(0, sqrt(tau));
}

mu = x * trees * weights;

lp += normal_lpdf(f | mu, a);
lp += normal_lpdf(weights | 0, sqrt(b));
lp += normal_lpdf(trees | 0, 1);

return lp;

}
}

SYNTAX ERROR, MESSAGE(S) FROM PARSER:
No matches for:

vector ~ bart_custom(int, int, real, int, vector)

Distribution bart_custom not found. Require function with _lpdf or _lpmf or _log suffix
Real return type required for probability function.
error in ‘modelb2b47f9297ad_BART_function’ at line 14, column 40

12: 
13: model {
14:   f ~ bart_custom_lpdf(0, 0, 0.5, 2, x); // BART prior
                                           ^
15:   y ~ normal(f, sigma);       // likelihood

Error in stanc(filename, allow_undefined = TRUE) :
failed to parse Stan model ‘BART function’ due to the above error.

2

Your functions block should be at the top of your Stan file (before the data block).

Additionally, you don’t need the _lpdf extension when using the sampling statement ~. So in the model block you should just call:

f ~ bart_custom(...)
3

Thank you very much. But I still get an error: SYNTAX ERROR, MESSAGE(S) FROM PARSER:
No matches for:

size(vector)

Available argument signatures for size:

size(int)
size(real)
size(vector)
size(row_vector)
size(matrix)
size(int[ , ])
size(real[ , ])
size(vector[ , ])
size(row_vector[ , ])
size(matrix[ , ])
size(int[ , , ])
size(real[ , , ])
size(vector[ , , ])
size(row_vector[ , , ])
size(matrix[ , , ])
size(int[ , , , ])
size(real[ , , , ])
size(vector[ , , , ])
size(row_vector[ , , , ])
size(matrix[ , , , ])
size(int[ , , , , ])
size(real[ , , , , ])
size(vector[ , , , , ])
size(row_vector[ , , , , ])
size(matrix[ , , , , ])
size(int[ , , , , , ])
size(real[ , , , , , ])
size(vector[ , , , , , ])
size(row_vector[ , , , , , ])
size(matrix[ , , , , , ])
size(int[ , , , , , , ])
size(real[ , , , , , , ])
size(vector[ , , , , , , ])
size(row_vector[ , , , , , , ])
size(matrix[ , , , , , , ])

error in ‘modelb2b43f2f648_BART_function’ at line 11, column 19

 9:     // x: predictor variables
10:     int num_trees = 50; // number of trees
11:     int N = size(x);
                      ^
12:     matrix[N, num_trees] trees;

Error in stanc(filename, allow_undefined = TRUE) :
failed to parse Stan model ‘BART function’ due to the above error.

4

rng functions can only be called in transformed data and generated quantities blocks. Also, any function that has a call to an rng function must itself be declared with as an rng function. The way this pdf is written cannot be estimated in Stan because of the above restrictions.

5

@spinkney I do not understand exactly where should I change. Can you please the lines?