inst/tests/test-config.R

In andrewzm/INLA: Functions which allow to perform full Bayesian analysis of latent Gaussian models using Integrated Nested Laplace Approximaxion

context("test 'config'")

test_that("Case 1", {
    set.seed(123)
    n = 100
    rho = 0.9
    x = arima.sim(n=n, model = list(ar = rho))
    x = x - mean(x)
    z = rnorm(n)
    y = 1 + z + x + rnorm(n, sd=0.01)
    
    idx = 1:n
    formula = y ~ 1 + z + f(idx, model="ar1", constr=TRUE)
    r = inla(formula,  data = data.frame(y, z, idx),
            control.compute = list(config = TRUE),
            control.inla = list(int.strategy = "grid", diff.logdens = 5, dz=.75))
    cs = r$misc$configs
    for (i in 1:length(cs$contents$tag)) {
        idx = seq(cs$contents$start[i],  len = cs$contents$len[i])
        mean = 0
        psum = 0
        for(k in 1:cs$nconfig) {
            mean = mean + exp(cs$config[[k]]$log.posterior) * cs$config[[k]]$mean[idx]
            psum = psum + exp(cs$config[[k]]$log.posterior)
        }
        mean = mean/psum
        desc = cs$contents$tag[i]
        if (desc == "z" || desc == "(Intercept)") {
            ##print(abs(mean - r$summary.fixed[desc, "mean"]))
            expect_true(abs(mean - r$summary.fixed[desc, "mean"]) < 0.001)
        }
    }
})