tests/testthat/test_overhead.R

# Benchmark the communication overhead in calling the algorithm
# multiple times for each iteration.
library(rstan)
rstan_options(auto_write = TRUE)
options(mc.cores = parallel::detectCores())
library(gmo)

data <- list(J = 8,
             K = 2,
             y = c(28,  8, -3,  7, -1,  1, 18, 12),
             sigma = c(15, 10, 16, 11,  9, 11, 10, 18))

local_file <- "models/8schools_local.stan"
M <- 1
draws <- 10
phi <- c(2,5)
seed <- 42

# TODO
# start with same alpha's
# Check conditional approximation.
g_alpha <- optimizing(stan_model(local_file),
                      data=c(data, list(phi=phi)),
                      seed=seed, #init=list(alpha=...), # TODO
                      draws=M*draws, constrained=FALSE)
                      #iter=5L) # TODO

for (t in 1:10) {
  g_alpha <- optimizing(stan_model(local_file),
                        data=c(data, list(phi=phi)),
                        seed=seed, #init=list(alpha=...), # TODO
                        draws=M*draws, constrained=FALSE)
                        iter=1L) # TODO
  # TODO
  # get alpha from previous iteration
}
stan-dev/gmo documentation built on May 26, 2017, 4:32 p.m.