We have External Comparer 6 after my thesis defense.

There were only minor changes from ExComp5

  1. We added BART, since it is popular.

  2. We added fit and prediction time. This is confusing for some that do fit and prediction simultaneously, such as BART. Or for aGP, which doesn't fit, it only predicts. Actually aGP has some premodel fitting, so it does have some fitting, but most of the time is prediction.

knitr::opts_chunk$set(echo = TRUE)
excomp <- readRDS("C:/Users/cbe117/Documents/GitHub/CGGP/scratch/ExternalComparison/ExComp6_almostall.rds")
excomp$plot_run_times()
# plyr::dlply(excomp$outcleandf, "d")
require('ggplot2');require('dplyr');require('magrittr');
table(excomp$completed_runs)

Details of failed runs:

# excomp$rungrid2()[!excomp$completed_runs,]
# All are beambending
ecdf <- excomp$outcleandf[excomp$completed_runs & !is.na(excomp$outcleandf$package),]
ecdf$n <- ecdf$npd * ecdf$d
ggplot(data=ecdf, mapping=aes(n, RMSE, color=package)) + geom_point() + facet_grid(f ~ package, scales="free_y") + scale_y_log10() + scale_x_log10()
ggplot(data=ecdf, mapping=aes(n, score, color=package)) + geom_point() + facet_grid(f ~ package, scales="free_y") + scale_x_log10()
ggplot(data=ecdf[ecdf$package!="mlegp",], mapping=aes(n, score, color=package)) + geom_point() + facet_grid(f ~ package, scales="free_y") + scale_x_log10()
ggplot(data=ecdf, mapping=aes(n, CRPscore)) + geom_point() + facet_grid(f ~ package, scales="free_y") + scale_y_log10()
ggplot(data=ecdf, mapping=aes(n, runtime)) + geom_point() + facet_grid(f ~ package, scales="free_y") + scale_y_log10() + scale_x_log10()
ggplot(data=ecdf, mapping=aes(n, fittime)) + geom_point() + facet_grid(f ~ package, scales="free_y") + scale_y_log10() + scale_x_log10()
ggplot(data=ecdf, mapping=aes(n, predtime)) + geom_point() + facet_grid(f ~ package, scales="free_y") + scale_y_log10() + scale_x_log10()
# saveRDS(excomp, "./scratch/ExternalComparison/ExComp1_completed.rds")
ggplot(data=ecdf %>% filter(package %in% c("CGGP","CGGPsupp", "CGGPoneshot")), mapping=aes(n, RMSE, color=correlation)) + geom_point() + facet_grid(f ~ interaction(package,correlation), scales="free_y") + scale_y_log10() + scale_x_log10()
ggplot(data=ecdf %>% filter(package %in% c("CGGP","CGGPsupp")), mapping=aes(n, RMSE, color=correlation)) + geom_point() + facet_grid(f ~ interaction(package,correlation), scales="free_y") + scale_y_log10() + scale_x_log10()
ggplot(data=ecdf, mapping=aes(n, RMSE, color=package)) + geom_point() + facet_grid(f ~ package, scales="free_y") + scale_y_log10() + scale_x_log10()


CollinErickson/CGGP documentation built on Feb. 6, 2024, 2:24 a.m.