Full model with t-distribution and no rainfall: burn-in and thinning

Load packages, read data and source custom scripts

rm(list = ls())
library(bamlss)

#> Loading required package: coda

#> Loading required package: colorspace

#> Loading required package: mgcv

#> Loading required package: nlme

#> This is mgcv 1.8-31. For overview type 'help("mgcv-package")'.

#> 
#> Attaching package: 'bamlss'

#> The following object is masked from 'package:mgcv':
#> 
#>     smooth.construct

library(gamlss.dist)

#> Loading required package: MASS

path_proj <- day2day::git_path()
path_data <- file.path(path_proj, "data")
path_processed <- file.path(path_data, "processed")
path_modelled <- file.path(path_data, "modelled")

bwdata_file <- file.path(path_processed, "bwdata_41_model.fst")
path_modelled_data <- file.path(path_modelled, "bw-15-full-no-rain-t.rds")
path_modelled_rectified <- gsub("(\\.rds)$", "-rectified\\1", path_modelled_data)
path_modelled_burned <- gsub("(\\.rds)$", "-burned\\1", path_modelled_data)
path_form <- gsub("(\\.rds)$", "-form\\1", path_modelled_data)

bwdata_model <- fst::read_fst(bwdata_file)
model <- readRDS(path_modelled_rectified)
form <- readRDS(path_form)

Burn-in and thinning

model$samples <- window(model$samples, start = 3001, thin = 20)

Compute results

system.time(model$results <- results.bamlss.default(model))

#>    user  system elapsed 
#>  53.957   0.199  54.157

system.time(saveRDS(model, file = path_modelled_burned))

#>    user  system elapsed 
#>   7.902   0.012   7.995

Maximum auto-correlation function (ACF)

par(mar = c(4, 4, 0.5, 0), mfrow = c(1, 3))
plot(model, model = "mu", which = "max-acf", spar = FALSE)
plot(model, model = "sigma", which = "max-acf", spar = FALSE)
plot(model, model = "nu", which = "max-acf", spar = FALSE)

Figure 1: Maximum ACF of samples for \(\mu\) (left), \(\sigma\) (center) and \(\nu\)(right)

MCMC convergence

par(mar = c(4, 4, 3, 1), mfrow = c(1, 2))
plot(model, which = "samples")

Time to execute the task

Only useful when executed with Rscript.

proc.time()

#>    user  system elapsed 
#>  95.082   0.569  95.817