compound declare/define and vectorization

[bob-carpenter]

I only looked at one of these models:

data {
  int<lower=1> n;
  vector[n] y;
}
parameters {
  # 確率的レベル
  vector[n] mu;
  # レベル撹乱項
  real<lower=0> sigma_level;
  # 観測撹乱項
  real<lower=0> sigma_irreg;
}
transformed parameters {
  vector[n] yhat;
  yhat = mu;
}
model {
  for(t in 2:n)
    mu[t] ~ normal(mu[t-1], sigma_level);

  y ~ normal(yhat, sigma_irreg);
}

and would like to make a few suggestions.

1. Use compound declare define wherever possible. Here, you can have a one-liner

vector[n] yhat = mu;

I'm not sure why you want to give this parameter vector two names. I'd suggest just removing the variable yhat altogether and sticking to mu.

2. Vectorize the time series model

mu[2:T] ~ normal(mu[1:T - 1], sigma_level);

3. Provide some kind of prior on mu[1],

mu[1] ~ normal(?, ??);

where the location and scale should be chosen based on problem knowledge.