add cmdstanr and make stan code more efficient

R/fit-cmdstanr.r

@@ -0,0 +1,61 @@
+#!/usr/bin/env Rscript
+## fit-cmdstanr.R
+## Bayesian MCMC fit using R-Stan
+
+if (!require("pacman")) install.packages("pacman")
+pacman::p_load("arrow", "smfsb", "cmdstanr")
+
+df = read_parquet(file.path("pima.parquet"))
+print(head(df))
+n = dim(df)[1]
+p = dim(df)[2]
+y = df[, p]
+y = as.integer(y)-1
+X = as.matrix(df[, -p])
+X = cbind(Int=1, X)
+print(y[1:6])
+print(head(X))
+
+set.seed(43) # fix seed to avoid choosing bad inits
+init = rnorm(p, 0.1)
+names(init) = colnames(X)
+
+modelstring ="
+data {
+  int<lower=1> n;
+  int<lower=1> p;
+  array[n] int<lower=0, upper=1> y;
+  matrix[n, p] X;
+}
+parameters {
+  vector[p] beta;
+}
+model {
+ beta[1] ~ normal(0, 10);
+ beta[2:p] ~ std_normal();
+ y ~ bernoulli_logit_glm(X[:, 2:p], beta[1], beta[2:p]);
+}
+"
+
+f <- write_stan_file(modelstring)
+constants = list(n=n, p=p, X=X, y=as.integer(y))
+thin = 2 # set thinning here
+
+print("Calling stan now...")
+mod <- cmdstan_model(f)
+
+mod_out <- mod$sample(
+  data = constants,
+  parallel_chains = 4,
+  iter_sampling = 2500 * thin + 1000,
+  iter_warmup = 1000,
+  thin = thin,
+  init = function() list(beta = init)
+)
+
+out <- mod_out$summary()
+out_draws <- mod_out$draws(format = "matrix")
+image(cor(out_draws))
+pairs(out_draws[sample(1:10000,1000),],pch=19,cex=0.2)
+## eof
+

R/fit-rstan.R

@@ -24,24 +24,17 @@ modelstring="
 data {
   int<lower=1> n;
   int<lower=1> p;
-  int<lower=0, upper=1> y[n];
-  real X[n,p];
+  array[n] int<lower=0, upper=1> y;
+  matrix[n, p] X;
 }
 parameters {
-  real beta[p];
+  vector[p] beta;
 }
 model {
-  for (i in 1:n) {
-    real eta = dot_product(beta, X[i,]);
-    real pr = 1/(1+exp(-eta));
-    y[i] ~ binomial(1, pr);
-  }
-  beta[1] ~ normal(0, 10);
-  for (j in 2:p) {
-    beta[j] ~ normal(0, 1);
-  }
+ beta[1] ~ normal(0, 10);
+ beta[2:p] ~ std_normal();
+ y ~ bernoulli_logit_glm(X[:, 2:p], beta[1], beta[2:p]);
 }
-
 "
 rstan_options(auto_write = TRUE)
 options(mc.cores = parallel::detectCores())