nile.stan file missing about example-models

[jrnold]

I don't remember actually adding this to the repo; to the extent I'm working on DLM stuff I'm working on a complete replacement. So I'd probably delete these examples before going back and figuring out how I did it.

[Emaasit]

@jrnold Do you have the current stan models for those two examples (nile & fx rates) somewhere else so I can follow along? I am trying to write dlm code for my work but there are limited examples on the internet.

[Emaasit]

I can't seem to find any examples for writing a generated quantities block for a Gaussian Dynamic Linear Model. I wrote the one shown below but stan produces an error. Any help will be much appreciated. @jrnold

stan_model1 <- "
// Kalman Filter (Multivariate Form)
// - no missing values
data {
int n_rows; // the number of rows/variables
int n_cols; // the number of columns/observations
matrix[n_rows, n_cols] y; // the observation matrix
vector[n_rows] mu_theta_init; // the mean at the initial state
cov_matrix[n_rows] sigma_theta_init; // covariance at the initial state

// holdout dataset
int n_rows_h;
int n_cols_h;
matrix[n_rows_h, n_cols_h] y_h;
vector[n_rows_h] mu_theta_init_h;
cov_matrix[n_rows_h] sigma_theta_init_h;
}
transformed data {
vector[n_rows] ones;
matrix[n_rows, n_rows] trans_matrix; // transition matrix
matrix[n_rows, n_rows] design_matrix; // design matrix
for (i in 1:n_rows) {
ones[i] <- 1.0;
}
trans_matrix <- diag_matrix(ones);
design_matrix <- trans_matrix;
}
parameters {
real<lower=-1.0, upper=1.0> rho;
vector<lower=0.0>[n_rows] sigma;
vector<lower=0.0>[n_rows] W_diag;
}
transformed parameters {
cov_matrix[n_rows] sigma_obs; //diagonal of observation covariance matrix of the y observations
cov_matrix[n_rows] sigma_theta; // state(system) covariance matrix of the states
sigma_theta <- diag_matrix(W_diag);
for (i in 1:n_rows) {
sigma_obs[i, i] <- pow(sigma[i], 2);
for (j in 1:(i - 1)) {
sigma_obs[i, j] <- sigma[i] * sigma[j] * rho;
sigma_obs[j, i] <- sigma_obs[i, j];
}
}
}
model {
y ~ gaussian_dlm_obs(design_matrix, trans_matrix, sigma_obs, sigma_theta, mu_theta_init, sigma_theta_init); // likelihood for gaussian dlm
}

generated quantities {
matrix[n_rows_h, n_cols_h] log_lik_h;

log_lik_h <- gaussian_dlm_obs_lpdf(y_h, design_matrix, trans_matrix, sigma_obs, sigma_theta, mu_theta_init_h, sigma_theta_init_h);
}
"

HERE IS THE ERROR MESSAGE:

ERROR at line 52

50: matrix[n_rows_h, n_cols_h] log_lik_h;
51:
52: log_lik_h <- gaussian_dlm_obs_lpdf(y_h, design_matrix, trans_matrix, sigma_obs, sigma_theta, mu_theta_init_h, sigma_theta_init_h);
^
53: }

PARSER EXPECTED:
Error in stanc(file = file, model_code = model_code, model_name = model_name, : failed to parse Stan model 'traffic-prediction1' due to the above error.

[bob-carpenter]

Would you mind moving this from this issue to our user lists? Only some of our developers regularly read issues and we want to keep the issue discussion focused on the issues being brought up.

[bob-carpenter]

The answer is that you need to use "|" in place of the first comma in an lpdf. I'm going to change that error message for the next release as per stan-dev/stan#1940

I'm closing this issue as it looks like @jrnold has an alternative fix in mind. Re-open if I'm closing in error.