In simisc/premiumPlots: Plot PReMiuM results and diagnostics

knitr::opts_chunk$set(echo = TRUE)
knitr::opts_knit$set(root.dir = knitr.root.directory)
# showwd <- function() {
#     rmdwd <- getwd()
#     stop(sprintf("now in %s", rmdwd))
# } 
# showwd() # looking for relative paths in PReMiuM/rmd (location of Rmd file), not in user's working directory

diss <- lapply(premium.models, calcDissimilarityMatrix)
clus <- lapply(diss, calcOptimalClustering) # choose maxClusters?
prof <- lapply(clus, calcAvgRiskAndProfile)

Choose top covariates

rho_ranks <- lapply(premium.models, function(m) {
  tibble(var = names(m$xMat),
         rank = rank(summariseVarSelectRho(m)$rhoMean))
  }) %>%
  bind_rows() %>%
  group_by(var) %>%
  summarise(rank = mean(rank)) %>%
  arrange(desc(rank))
(top_covariates <- rho_ranks$var[1:10])
(low_covariates <- rho_ranks$var[155:163])
do.call(getHyperparams, premium.models)

Plot summaries

do.call(plotVarSelectRho, prof)
do.call(plotClusterSizes, prof)
do.call(plotSimilarityMatrix, diss)

Plot profiles

lapply(prof, plotResponse)
lapply(prof, plotCovariateProfiles, whichCovariates = top_covariates) # why warnings?
lapply(prof, plotCovariateProfiles, whichCovariates = low_covariates) # using starred profile

Plot diagnostics

alpha.mcmc <- do.call(codaFromPremium, c("alpha", premium.models))
summary(alpha.mcmc)
plot(alpha.mcmc)
par(mfrow = c(3, 1))
autocorr.plot(alpha.mcmc, auto.layout = F)
gelman.plot(alpha.mcmc)
nclus.mcmc <- do.call(codaFromPremium, c("nClusters", premium.models))
summary(nclus.mcmc)
plot(nclus.mcmc)
par(mfrow = c(3, 1))
autocorr.plot(nclus.mcmc, auto.layout = F)
gelman.plot(nclus.mcmc)