inst/reproduce/biaseddata.posterior2given1.run.R

In pierrejacob/bettertogether: Better Together? Joint model versus moduarlized approach

library(bettertogether)
rm(list = setdiff(setdiff(ls(), "scriptfolder"), "resultsfolder"))
setmytheme()
registerDoParallel(cores = detectCores())
set.seed(16)

## data
load("biaseddata_dataset.RData")

y1 <- matrix(synthetic_dataset$y1, ncol = 1)
y2 <- matrix(synthetic_dataset$y2, ncol = 1)
#
### Model
## Module 1
module1 <- biaseddata_module1()
hyper1 <- module1$hyper1
## Module 2
module2 <- biaseddata_module2()
hyper2 <- module2$hyper2
#
hyper1$theta1_prior_sd <- 1
hyper2$theta2_prior_sd <- 0.1
## Algorithmic parameters
param_algo <- list(nthetas = 1024, minimum_diversity = 0.8, nmoves = 10, proposal = mixture_rmixmod())
#
# whether to compute things or load pre-computed things
doRun <- TRUE
rep <- 5
###
### Posterior in Module 1 given Module 1
filename <- paste0("biaseddata_module1.N", param_algo$nthetas, ".K", param_algo$nmoves, ".rep", rep, ".RData")
load(filename)
### and now assimilate Y2
target2given1 <- list(thetadim = 2, ydim = 1,
                      dprior = function(thetas, ...){
                        dnorm(thetas[,1], mean = hyper1$theta1_mean_prior, sd = hyper1$theta1_prior_sd, log = TRUE) +
                          dnorm(thetas[,2], mean = hyper2$theta2_mean_prior, sd = hyper2$theta2_prior_sd, log = TRUE) +
                          module1$loglik(theta1s = thetas[,1], y1 = y1)
                      },
                      fullloglikelihood = function(thetas, ys, ...){
                        return(module2$loglik(thetas[,1], thetas[,2], ys[,1]))
                      },
                      conditionallikelihood = function(thetas, ys, idata, ...){
                        return(module2$loglik(thetas[,1], thetas[,2], ys[idata,1]))
                      },
                      parameters = list())


filename <- paste0("biaseddata_module2givenmodule1.N", param_algo$nthetas, ".K", param_algo$nmoves, ".rep", rep, ".RData")
if (doRun){
  results2given1 <- foreach(i = 1:rep) %dorng% {
    thetas1 <- results1[[i]]$thetas
    normw1 <- results1[[i]]$normw
    thetas <- cbind(thetas1, rnorm(nrow(thetas1), mean = hyper2$theta2_mean_prior, sd = hyper2$theta2_prior_sd))
    res <- smcsampler(y2, target2given1, param_algo, thetas, normw1)
    list(thetas = res$thetas_history[[n2+1]], normw = res$normw_history[[n2+1]], logevidence = res$logevidence)
  }
  save(results2given1, file = filename)
} else {
  load(file = filename)
}