interactive_scripts/temp_kfold.R

In Famondir/birtms: Creates and analyses Bayesian IRT models

library(future)
<<<<<<< HEAD
options(mc.cores=10)
plan(multisession)

fit_1d_1pl_spm <- readRDS("../models/gdcp/fit_1d_1pl_spm_full1b.rds")
=======
options(mc.cores=4)
plan(multisession)

fit_1d_1pl_spm_full1 <- readRDS("../models/gdcp/fit_1d_1pl_spm_full1b.rds")
>>>>>>> 53ecf2d73a81e7cc384f7fac04e660d47da3dd72

# kfol1 + kfold2 brauchten 6:17 min
# kfold1 <- brms::kfold(fit_1d_1pl_spm_full1, chains = 1)
# kfold2 <- brms::kfold(fit_1d_1pl_spm_full1, chains = 1,
#                       folds = "grouped", group = "person")
kfold3 <- brms::kfold(fit_1d_1pl_spm_full1, chains = 1,
                      folds = "grouped", group = "person",
                      # K = length(unique(fit_1d_1pl_spm$data$person)))
<<<<<<< HEAD
                      K = 20)
=======
                      K = 80)
>>>>>>> 53ecf2d73a81e7cc384f7fac04e660d47da3dd72
# 150 Modelle in 1 h; nach 16 h bei 466 von 499 Modellen
# 471 nach 16.5 h
# Arbeitsspeicher vollgelaufen

<<<<<<< HEAD
saveRDS(kfold3, "../models/gdcp/kfold3.rds")
=======
saveRDS(kfold3, "../models/gdcp/80fold.rds")
>>>>>>> 53ecf2d73a81e7cc384f7fac04e660d47da3dd72
#
# loo1 <- brms::loo(fit_1d_1pl_spm_full1)
# # loo_marginal brauchte 40 s
# loo2 <- birtms::loo_marginal(fit_1d_1pl_spm_full1)
#
# kfold_1PL_cond <- kfold1
# kfold_1PL_marg <- kfold2
# loo_1PL_cond <- loo1
# loo_1PL_marg <- loo2
#
# ic_1pl <- list(kfold_1PL_cond, kfold_1PL_marg, loo_1PL_cond, loo_1PL_marg)
#
# saveRDS(ic_1pl, "models/gdcp/ic_1pl_spm.rds")