Multimodal model with a neat coding style

data.json

{
  "D": 2,
  "r": 2.5,
  "p": 0.6666666666666666
}

data.py

D = 2
r = 2.5
p = 2.0 / 3.0
data = {'D': D, 'r': r, 'p': p}

main.stan

functions {
  real mm_lpdf(vector y, real r, real p, int D) {
    real lp = 0;
    for (d in 1 : D) {
      real lp1 = lp + normal_lupdf(y[d] | r, 1);
      real lp2 = lp + normal_lupdf(y[d] | -r, 1);
      lp = log_mix(p, lp1, lp2);
    }
    return lp;
  }
}
data {
  int<lower=0> D; // number of dimensions
  real<lower=0> r; // modes in {-r, r}^D
  real<lower=0, upper=1> p; // p = Pr[Y[d] > 0]
}
parameters {
  vector[D] y;
}
model {
  y ~ mm(r, p, D);
}

meta.json

{"title":"Multimodal","dataSource":"generated_by_python"}

sampling_opts.json

{"num_chains":4,"num_warmup":1000,"num_samples":1000,"init_radius":2}