scratch/ExternalComparison/ExternalComparer4.R
In CollinErickson/CGGP: Composite Grid Gaussian Processes
# Comparison for paper
# decentLHS <- sFFLHD::decentLHS
decentLHS <- function(n, d, ndes, max.time) {
if (missing(ndes) && missing(max.time)) {
stop("Must give ndes or max.time to decentLHS")
}
# MaxProMeasure in 10d with 3e4 pts takes 3 min, crashes with 1e5 pts.
if (n > 1e4) {
return(lhs::randomLHS(n, d))
}
start.time <- Sys.time()
bestdes <- NULL
bestcrit <- Inf
i <- 1
while (TRUE) {
# Make new LHS
x <- lhs::randomLHS(n, d)
crit <- MaxPro::MaxProMeasure(x)
# Check if best yet
if (crit < bestcrit) {
bestcrit <- crit
bestdes <- x
}
# Check if done
i <- i+1
if (!missing(ndes) && (i > ndes)) {
break
}
if (!missing(max.time) && (as.numeric(Sys.time() - start.time, units="secs") > max.time)) {
break
}
}
# Return bestdes
bestdes
}
run_lagp <- function(Ntotal, Nappend, f, d, x, y, xtest, ytest, seed, use_agp=FALSE) {
if (!missing(seed)) {set.seed(seed)}
if (missing(x) && missing(y)) {
if (Ntotal<=2000) {x <- lhs::maximinLHS(Ntotal, d)}
else {x <- decentLHS(Ntotal, d, max.time=15)}
y <- apply(x, 1, f)
}
sdy <- sd(y)
mny <- mean(y)
y <- (y-mny) / sdy
if (use_agp) {
stop("Don't use this version")
fit.time.start <- Sys.time()
fit.time.end <- Sys.time()
pred.time.start <- Sys.time()
pred <- laGP::aGPsep(X=x, Z=y, XX=xtest, method="alc")
pred$var <- pred$var * sdy^2
} else {
fit.time.start <- Sys.time()
mod.agp <- laGP::newGPsep(X=x, Z=y, d=laGP::darg(d=list(mle = TRUE, max = 100), X=x)$start,
g=laGP::garg(g=list(mle = TRUE), y=y)$start)
laGP::updateGPsep(mod.agp, x, y)
fit.time.end <- Sys.time()
pred.time.start <- Sys.time()
pred <- laGP::predGPsep(mod.agp, xtest, lite=T)
pred$var <- pred$s2 * sdy^2
}
# browser()
pred$mean <- pred$mean * sdy + mny
pred.time.end <- Sys.time()
list(mean=pred$mean, var=pred$var, n=nrow(x),
pred.time=as.numeric(pred.time.end - pred.time.start, units="secs"),
fit.time =as.numeric(fit.time.end - fit.time.start , units="secs"))
}
run_lagp_bobby <- function(Ntotal, Nappend, f, d, x, y, xtest, ytest, seed, use_agp=FALSE) {
require(laGP)
if (!missing(seed)) {set.seed(seed)}
if (missing(x) && missing(y)) {
if (Ntotal<=2000) {x <- lhs::maximinLHS(Ntotal, d)}
else {x <- decentLHS(Ntotal, d, max.time=15)}
y <- apply(x, 1, f)
}
fit.time.start <- Sys.time()
## fixing a tiny nugget is very helpful on this problem
g <- 1/10000000
## macro-scale analysis on a random subset of the data
# browser()
n <- min(Ntotal, 1000)
d2 <- darg(list(mle = TRUE, max = 100), x)
subs <- sample(1:Ntotal, n, replace = FALSE)
gpsepi <- newGPsep(x[subs, ], y[subs], rep(d2$start, d), g=g, dK=TRUE)
that <- mleGPsep(gpsepi, param="d", tmin=d2$min, tmax=d2$max, ab=d2$ab, maxit=200)
# p <- predGPsep(gpsepi, xpred, lite=TRUE)
# rmse.sub[r] <- sqrt(mean((p$mean - ypred.0)^2))
deleteGPsep(gpsepi)
## scale the inputs according to the macro-analysis lengthscales
scale <- sqrt(that$d)
xs <- x
xtests <- xtest
for(j in 1:ncol(xs)) {
xs[,j] <- xs[,j] / scale[j]
xtests[,j] <- xtests[,j] / scale[j]
}
fit.time.end <- Sys.time()
pred.time.start <- Sys.time()
## laGP analysis on scaled inputs distributed over 8 cores.
## If you don't have 8 cores, change omp.threads to something smaller
# out <- aGP(xs, y, xtests, d=list(start=1, max=20), g=g, verb=0)
out.sep <- aGPsep(xs, y, xtests, d=list(start=1, max=20), g=g, verb=0,
end= min(50, Ntotal-1))
pred.time.end <- Sys.time()
# browser()
list(mean=out.sep$mean, var=out.sep$var, n=nrow(xs),
pred.time=as.numeric(pred.time.end - pred.time.start, units="secs"),
fit.time =as.numeric(fit.time.end - fit.time.start , units="secs"))
}
run_MRFA <- function(Ntotal, Nappend, f, d, x, y, xtest, ytest, seed) {
if (!missing(seed)) {set.seed(seed)}
if (missing(x) && missing(y)) {
if (Ntotal<=2000) {x <- lhs::maximinLHS(Ntotal, d)}
else {x <- decentLHS(Ntotal, d, max.time=15)}
y <- apply(x, 1, f)
}
fit.time.start <- Sys.time()
mod <- MRFA::MRFA_fit(X=x, Y=y, verbose=FALSE)
fit.time.end <- Sys.time()
pred.time.start <- Sys.time()
pred <- predict(mod, xtest, lambda = min(mod$lambda))
pred.time.end <- Sys.time()
list(mean=pred$y_hat, var=rep(NaN, nrow(xtest)), n=nrow(x),
pred.time=as.numeric(pred.time.end - pred.time.start, units="secs"),
fit.time =as.numeric(fit.time.end - fit.time.start , units="secs"))
}
run_svm <- function(Ntotal, Nappend, f, d, x, y, xtest, ytest, seed) {#browser()
if (!missing(seed)) {set.seed(seed)}
if (missing(x) && missing(y)) {
if (Ntotal<=2000) {x <- lhs::maximinLHS(Ntotal, d)}
else {x <- decentLHS(Ntotal, d, max.time=15)}
y <- apply(x, 1, f)
}
require(e1071)
fit.time.start <- Sys.time()
mod <- svm(x, y)
fit.time.end <- Sys.time()
pred.time.start <- Sys.time()
pred <- predict(mod, xtest)
pred.time.end <- Sys.time()
list(mean=pred, var=rep(NaN, nrow(xtest)), n=nrow(x),
pred.time=as.numeric(pred.time.end - pred.time.start, units="secs"),
fit.time =as.numeric(fit.time.end - fit.time.start , units="secs"))
}
run_mlegp <- function(Ntotal, Nappend, f, d, x, y, xtest, ytest, seed) {#browser()
if (!missing(seed)) {set.seed(seed)}
if (missing(x) && missing(y)) {
if (Ntotal<=2000) {x <- lhs::maximinLHS(Ntotal, d)}
else {x <- decentLHS(Ntotal, d, max.time=15)}
y <- apply(x, 1, f)
}
require(mlegp)
fit.time.start <- Sys.time()
mod <- mlegp(x, y)
fit.time.end <- Sys.time()
pred.time.start <- Sys.time()
pred <- predict(mod, xtest, se.fit=T)
pred.time.end <- Sys.time()
list(mean=pred$fit, var=pred$se.fit^2, n=nrow(x),
pred.time=as.numeric(pred.time.end - pred.time.start, units="secs"),
fit.time =as.numeric(fit.time.end - fit.time.start , units="secs"))
}
run_GPfit <- function(Ntotal, Nappend, f, d, x, y, xtest, ytest, seed) {#browser()
if (!missing(seed)) {set.seed(seed)}
if (missing(x) && missing(y)) {
if (Ntotal<=2000) {x <- lhs::maximinLHS(Ntotal, d)}
else {x <- decentLHS(Ntotal, d, max.time=15)}
y <- apply(x, 1, f)
}
require(GPfit)
fit.time.start <- Sys.time()
mod <- GP_fit(x, y)
fit.time.end <- Sys.time()
pred.time.start <- Sys.time()
pred <- predict(mod, xtest)
pred.time.end <- Sys.time()
list(mean=pred$Y_hat, var=pred$MSE, n=nrow(x),
pred.time=as.numeric(pred.time.end - pred.time.start, units="secs"),
fit.time =as.numeric(fit.time.end - fit.time.start , units="secs"))
}
run_CGGP <- function(Ntotal, Nappend, Nlhs, f, d, x, y, xtest, ytest, seed, selection.method, correlation) {#browser()
require("CGGP")
if (!missing(seed)) {set.seed(seed)}
if (!missing(Nlhs) && Nlhs!=0) {stop("Nlhs given to run_CGGP")}
fit.time.start <- Sys.time()
sg <- CGGPcreate(d=d, batchsize=min(Ntotal,200), corr=correlation)
sg <- CGGPfit(sg, apply(sg$design, 1, f))
notdone <- (nrow(sg$design) < Ntotal)
while (notdone) {
# print(sg)
Nalready <- nrow(sg$design)
ni <- if (Nalready<1000) {200} else if (Nalready<10000) {500} else if (Nalready<20000) {2000} else {10000}
if (ni +Nalready > Ntotal) {
ni <- Ntotal - Nalready
notdone <- FALSE
}
sg <- CGGPappend(sg, batchsize = ni, selectionmethod = selection.method)
if (!is.null(sg$design_unevaluated)) {
ynew <- apply(sg$design_unevaluated, 1, f)
sg <- CGGPfit(sg, Ynew=ynew)
} else {
print('Nothing new to evaluate, hopefully !notdone')
}
}
fit.time.end <- Sys.time()
# print(sg)
Nevaluated <- if (!is.null(sg$design)) nrow(sg$design) else {0} + nrow(sg$Xs)
if (Nevaluated > Ntotal) {stop("CGGP has more points than Ntotal")}
pred.time.start <- Sys.time()
pred <- predict(sg, xtest)
pred.time.end <- Sys.time()
list(mean=pred$mean, var=pred$var,
n=Nevaluated,
pred.time=as.numeric(pred.time.end - pred.time.start, units="secs"),
fit.time =as.numeric(fit.time.end - fit.time.start , units="secs"))
}
run_CGGPsupp <- function(Ntotal, Nappend, Nlhs, f, d, x, y, xtest, ytest, seed, HandlingSuppData, selection.method, correlation) {#browser()
require("CGGP")
if (!missing(seed)) {set.seed(seed)}
xsup <- lhs::maximinLHS(Nlhs, d)
ysup <- apply(xsup, 1, f)
fit.time.start <- Sys.time()
sg <- CGGPcreate(d=d, batchsize=0, Xs=xsup, Ys=ysup, corr=correlation)
notdone <- (Nlhs < Ntotal)
while (notdone) {
# print(sg)
Nalready <- (if (!is.null(sg$design)) {nrow(sg$design)} else {0}) + nrow(sg$Xs)
ni <- if (Nalready<1000) {200} else if (Nalready<10000) {500} else if (Nalready<20000) {2000} else {10000}
if (ni +Nalready > Ntotal) {
ni <- Ntotal - Nalready
notdone <- FALSE
}
sg <- CGGPappend(sg, batchsize = ni, selection.method)
if (!is.null(sg$design_unevaluated)) {
ynew <- apply(sg$design_unevaluated, 1, f)
sg <- CGGPfit(sg, Ynew=ynew, Xs=xsup, Ys=ysup, HandlingSuppData = HandlingSuppData)
} else {
print('Nothing new to evaluate, hopefully !notdone')
}
}
fit.time.end <- Sys.time()
# print(sg)
Nevaluated <- if (!is.null(sg$design)) nrow(sg$design) else {0} + nrow(sg$Xs)
if (Nevaluated > Ntotal) {stop("CGGP has more points than Ntotal")}
pred.time.start <- Sys.time()
pred <- predict(sg, xtest)
pred.time.end <- Sys.time()
list(mean=pred$mean, var=pred$var,
n=Nevaluated,
pred.time=as.numeric(pred.time.end - pred.time.start, units="secs"),
fit.time =as.numeric(fit.time.end - fit.time.start , units="secs"))
}
run_CGGPsupponly <- function(Ntotal, Nappend, Nlhs, f, d, x, y, xtest, ytest, seed, correlation) {#browser()
require("CGGP")
if (!missing(seed)) {set.seed(seed)}
if (Ntotal > 2000) {stop("CGGPsupponly can't run with more than 2000")}
xsup <- lhs::maximinLHS(Ntotal, d)
ysup <- apply(xsup, 1, f)
fit.time.start <- Sys.time()
sg <- CGGPcreate(d=d, batchsize=0, Xs=xsup, Ys=ysup, corr=correlation)
fit.time.end <- Sys.time()
# print(sg)
Nevaluated <- nrow(sg$Xs)
if (Nevaluated > Ntotal) {stop("CGGP has more points than Ntotal")}
pred.time.start <- Sys.time()
pred <- predict(sg, xtest)
pred.time.end <- Sys.time()
list(mean=pred$mean, var=pred$var,
n=Nevaluated,
pred.time=as.numeric(pred.time.end - pred.time.start, units="secs"),
fit.time =as.numeric(fit.time.end - fit.time.start , units="secs"))
}
run_CGGPoneshot <- function(Ntotal, Nappend, Nlhs, f, d, x, y, xtest, ytest, seed, correlation) {#browser()
require("CGGP")
if (!missing(seed)) {set.seed(seed)}
# xsup <- lhs::maximinLHS(Nlhs, d)
# ysup <- apply(xsup, 1, f)
fit.time.start <- Sys.time()
sg <- CGGPcreate(d=d, batchsize=Ntotal, corr=correlation)
sg <- CGGPfit(sg, Y=apply(sg$design, 1, f))
fit.time.end <- Sys.time()
# print(sg)
Nevaluated <- nrow(sg$design)
if (Nevaluated > Ntotal) {stop("CGGP has more points than Ntotal")}
pred.time.start <- Sys.time()
pred <- predict(sg, xtest)
pred.time.end <- Sys.time()
list(mean=pred$mean, var=pred$var,
n=Nevaluated,
pred.time=as.numeric(pred.time.end - pred.time.start, units="secs"),
fit.time =as.numeric(fit.time.end - fit.time.start , units="secs"))
}
# Need a generic function that passes to specific ones
run_one <- function(package, selection.method, correlation, HandlingSuppData,
f, d, npd, replicate) {#browser()
# package <- psch$package
n <- npd * d
# if (n!= 500) {stop('bad n')}
f <- eval(parse(text=paste0("TestFunctions::", f)))
ntest <- 1e4
# xtest <- matrix(runif(ntest*d), ncol=d)
xtest <- decentLHS(ntest, d, max.time = 10)
ytest <- apply(xtest, 1, f)
# if (package == "CGGP") {
# out <- run_CGGP(Nappend=floor(n * (1:5)/5), f=f, d=d, xtest=xtest)
if (package == "CGGPsupp") {
# out <- run_CGGP(Nappend=floor(n*(2:5)/5), Nlhs=floor(.2*n), f=f, d=d, xtest=xtest)
out <- run_CGGPsupp(Ntotal=n, Nlhs=10*d, f=f, d=d, xtest=xtest, HandlingSuppData=HandlingSuppData, selection.method=selection.method, correlation=correlation)
# } else if (package == "CGGPsupponly") {
# out <- run_CGGP(Nappend=c(), Nlhs=n, f=f, d=d, xtest=xtest)
} else if (package == "CGGPsupponly") {
out <- run_CGGPsupponly(Ntotal=n, f=f, d=d, xtest=xtest, correlation=correlation)
} else if (package == "CGGPoneshot") {
out <- run_CGGPoneshot(Ntotal=n, f=f, d=d, xtest=xtest, correlation=correlation)
} else if (package == "CGGP") {
out <- run_CGGP(Ntotal=n, f=f, d=d, xtest=xtest, selection.method=selection.method, correlation=correlation)
} else if (package == "laGP") {
out <- run_lagp(Ntotal=n, f=f, d=d, xtest=xtest)
} else if (package == "aGP") {
out <- run_lagp_bobby(Ntotal=n, f=f, d=d, xtest=xtest, use_agp=TRUE)
} else if (package == "MRFA") {
out <- run_MRFA(Ntotal=n, f=f, d=d, xtest=xtest)
} else if (package == "svm") {
out <- run_svm(Ntotal=n, f=f, d=d, xtest=xtest)
} else if (package == "mlegp") {
out <- run_mlegp(Ntotal=n, f=f, d=d, xtest=xtest)
} else if (package == "GPfit") {
out <- run_GPfit(Ntotal=n, f=f, d=d, xtest=xtest)
} else {
stop(paste("Package", package, "not recognized"))
}
# browser()
outstats <- CGGP::valstats(predmean=out[[1]], predvar=out[[2]],Yval=ytest) #, bydim=FALSE)
if (out$n > n) {warning(paste("n too big for", package, n, f, d))}
outstats
}
expand.grid.df <- function(...) Reduce(function(...) merge(..., by=NULL), list(...))
eg1 <- expand.grid(selection.method = c("UCB", "Greedy"),
correlation = c("CauchySQ", "Matern32", "PowerExp"), stringsAsFactors=F)
eg2a <- expand.grid.df(eg1, data.frame(HandlingSuppData=c("Ignore", "Correct"), stringsAsFactors=F), data.frame(package="CGGPsupp", stringsAsFactors=F))
eg2b <- expand.grid.df(eg1, data.frame(HandlingSuppData="NA", stringsAsFactors=F), data.frame(package=c("CGGP"), stringsAsFactors=F))
eg2c <- expand.grid(selection.method="NA", correlation = c("CauchySQ", "Matern32", "PowerExp"), HandlingSuppData="NA", package=c("CGGPoneshot", "CGGPsupponly"), stringsAsFactors=F)
eg2d <- data.frame(selection.method="NA", correlation="NA", HandlingSuppData="NA",
package=c("MRFA", "svm", "aGP", "laGP", "mlegp", "GPfit"), stringsAsFactors=F)
eg3 <- rbind(eg2a, eg2b, eg2c, eg2d)
require("comparer")
excomp <- ffexp$new(
eval_func = run_one,
varlist = c("decentLHS", "run_CGGP", "run_CGGPoneshot", "run_CGGPsupp",
"run_CGGPsupponly", "run_GPfit", "run_lagp", "run_lagp_bobby",
"run_mlegp", "run_MRFA", "run_svm"),
fd=data.frame(f=c("beambending","OTL_Circuit","piston","borehole","wingweight"),
d=c(3,6,7,8,10),
row.names = c("beam","OTL","piston","borehole","wingweight"), stringsAsFactors = F),
# package=c("CGGPsupp", "CGPPoneshot", "CGGPsupponly", "CGGP",
# "MRFA", "svm", "aGP", "laGP", "mlegp", "GPfit"),
psch=eg3,
npd=c(10, 30, 100, 300, 1000, 3000, 10000),
parallel=TRUE,
parallel_cores = 20,
replicate=1:10, #:5,
folder_path= "/home/collin/scratch/SGGP/scratch/ExternalComparison/ExComp4"
# folder_path="./scratch/ExternalComparison/ExComp4/"
)
# Remove ones that can't do full size
package.name <- excomp$arglist$psch$package[excomp$rungrid$psch]
npd.excomp <- excomp$arglist$npd[excomp$rungrid$npd]
n.excomp <- npd.excomp * excomp$arglist$fd$d[excomp$rungrid$fd]
excomp$completed_runs[package.name == "CGGPsupponly" & n.excomp > 1000] <- TRUE
excomp$completed_runs[package.name == "laGP" & n.excomp > 1000] <- TRUE
excomp$completed_runs[package.name == "mlegp" & n.excomp > 400] <- TRUE
excomp$completed_runs[package.name == "GPfit" & n.excomp > 100] <- TRUE
# agp is giving errors
excomp$completed_runs[package.name == "aGP"] <- TRUE
table(paste(package.name, excomp$completed_runs))
# plyr::ddply(data.frame(package.name, compruns=excomp$completed_runs), "package.name", function(d) {data.frame(done=sum(d$compruns), notdone=sum(!d$compruns))})
rbind(plyr::ddply(data.frame(package.name, compruns=excomp$completed_runs), "package.name", function(d) {data.frame(done=sum(d$compruns), notdone=sum(!d$compruns))}), data.frame(package.name="all", done=sum(excomp$completed_runs), notdone=sum(!excomp$completed_runs)))
# excomp$run_one(1752)
# excomp$run_one(7299)
# excomp$run_all(save_output = T, parallel = F, parallel_temp_save = T, run_order = "random")
# excomp$rungrid
# try because it gave delete error before, but shouldn't need it now
table(excomp$completed_runs)
try(excomp$recover_parallel_temp_save(delete_after = FALSE))
table(excomp$completed_runs)
# plyr::ddply(data.frame(package.name, compruns=excomp$completed_runs), "package.name", function(d) {data.frame(done=sum(d$compruns), notdone=sum(!d$compruns))})
rbind(plyr::ddply(data.frame(package.name, compruns=excomp$completed_runs), "package.name", function(d) {data.frame(done=sum(d$compruns), notdone=sum(!d$compruns))}), data.frame(package.name="all", done=sum(excomp$completed_runs), notdone=sum(!excomp$completed_runs)))
excomp$save_self()
# excomp$run_all(parallel_temp_save = TRUE, delete_parallel_temp_save_after=FALSE,
# write_start_files=T, write_error_files=T)
# Getting errors, run by package.name to see which is causing it
excomp$parallel_cores <- 10
excomp$run_all(to_run = which(!excomp$completed_runs & (package.name == "mlegp")),
parallel_temp_save = TRUE, delete_parallel_temp_save_after=FALSE,
write_start_files=T, write_error_files=T)
# excomp$run_all()
cat("Completed all runs in ExternalComparer4.R\n")
excomp$save_self()
cat("Saved self\n")
if (F) {
excomp <- readRDS("./scratch/ExternalComparison/ExternalComparer4_mostlydone.rds")
excompbass <- readRDS("./scratch/ExternalComparison/ExternalComparer4bass_quarterdone.rds")
excomp$plot_run_times()
plyr::dlply(excomp$outcleandf, "d")
require('ggplot2')
ecdf <- rbind(excomp$outcleandf[excomp$completed_runs & !is.na(excomp$outcleandf$package),],
excompbass$outcleandf[excompbass$completed_runs & !is.na(excompbass$outcleandf$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()
# saveRDS(excomp, "./scratch/ExternalComparison/ExComp1_completed.rds")
}
CollinErickson/CGGP documentation built on Feb. 6, 2024, 2:24 a.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
# Comparison for paper
# decentLHS <- sFFLHD::decentLHS
decentLHS <- function(n, d, ndes, max.time) {
if (missing(ndes) && missing(max.time)) {
stop("Must give ndes or max.time to decentLHS")
}
# MaxProMeasure in 10d with 3e4 pts takes 3 min, crashes with 1e5 pts.
if (n > 1e4) {
return(lhs::randomLHS(n, d))
}
start.time <- Sys.time()
bestdes <- NULL
bestcrit <- Inf
i <- 1
while (TRUE) {
# Make new LHS
x <- lhs::randomLHS(n, d)
crit <- MaxPro::MaxProMeasure(x)
# Check if best yet
if (crit < bestcrit) {
bestcrit <- crit
bestdes <- x
}
# Check if done
i <- i+1
if (!missing(ndes) && (i > ndes)) {
break
}
if (!missing(max.time) && (as.numeric(Sys.time() - start.time, units="secs") > max.time)) {
break
}
}
# Return bestdes
bestdes
}
run_lagp <- function(Ntotal, Nappend, f, d, x, y, xtest, ytest, seed, use_agp=FALSE) {
if (!missing(seed)) {set.seed(seed)}
if (missing(x) && missing(y)) {
if (Ntotal<=2000) {x <- lhs::maximinLHS(Ntotal, d)}
else {x <- decentLHS(Ntotal, d, max.time=15)}
y <- apply(x, 1, f)
}
sdy <- sd(y)
mny <- mean(y)
y <- (y-mny) / sdy
if (use_agp) {
stop("Don't use this version")
fit.time.start <- Sys.time()
fit.time.end <- Sys.time()
pred.time.start <- Sys.time()
pred <- laGP::aGPsep(X=x, Z=y, XX=xtest, method="alc")
pred$var <- pred$var * sdy^2
} else {
fit.time.start <- Sys.time()
mod.agp <- laGP::newGPsep(X=x, Z=y, d=laGP::darg(d=list(mle = TRUE, max = 100), X=x)$start,
g=laGP::garg(g=list(mle = TRUE), y=y)$start)
laGP::updateGPsep(mod.agp, x, y)
fit.time.end <- Sys.time()
pred.time.start <- Sys.time()
pred <- laGP::predGPsep(mod.agp, xtest, lite=T)
pred$var <- pred$s2 * sdy^2
}
# browser()
pred$mean <- pred$mean * sdy + mny
pred.time.end <- Sys.time()
list(mean=pred$mean, var=pred$var, n=nrow(x),
pred.time=as.numeric(pred.time.end - pred.time.start, units="secs"),
fit.time =as.numeric(fit.time.end - fit.time.start , units="secs"))
}
run_lagp_bobby <- function(Ntotal, Nappend, f, d, x, y, xtest, ytest, seed, use_agp=FALSE) {
require(laGP)
if (!missing(seed)) {set.seed(seed)}
if (missing(x) && missing(y)) {
if (Ntotal<=2000) {x <- lhs::maximinLHS(Ntotal, d)}
else {x <- decentLHS(Ntotal, d, max.time=15)}
y <- apply(x, 1, f)
}
fit.time.start <- Sys.time()
## fixing a tiny nugget is very helpful on this problem
g <- 1/10000000
## macro-scale analysis on a random subset of the data
# browser()
n <- min(Ntotal, 1000)
d2 <- darg(list(mle = TRUE, max = 100), x)
subs <- sample(1:Ntotal, n, replace = FALSE)
gpsepi <- newGPsep(x[subs, ], y[subs], rep(d2$start, d), g=g, dK=TRUE)
that <- mleGPsep(gpsepi, param="d", tmin=d2$min, tmax=d2$max, ab=d2$ab, maxit=200)
# p <- predGPsep(gpsepi, xpred, lite=TRUE)
# rmse.sub[r] <- sqrt(mean((p$mean - ypred.0)^2))
deleteGPsep(gpsepi)
## scale the inputs according to the macro-analysis lengthscales
scale <- sqrt(that$d)
xs <- x
xtests <- xtest
for(j in 1:ncol(xs)) {
xs[,j] <- xs[,j] / scale[j]
xtests[,j] <- xtests[,j] / scale[j]
}
fit.time.end <- Sys.time()
pred.time.start <- Sys.time()
## laGP analysis on scaled inputs distributed over 8 cores.
## If you don't have 8 cores, change omp.threads to something smaller
# out <- aGP(xs, y, xtests, d=list(start=1, max=20), g=g, verb=0)
out.sep <- aGPsep(xs, y, xtests, d=list(start=1, max=20), g=g, verb=0,
end= min(50, Ntotal-1))
pred.time.end <- Sys.time()
# browser()
list(mean=out.sep$mean, var=out.sep$var, n=nrow(xs),
pred.time=as.numeric(pred.time.end - pred.time.start, units="secs"),
fit.time =as.numeric(fit.time.end - fit.time.start , units="secs"))
}
run_MRFA <- function(Ntotal, Nappend, f, d, x, y, xtest, ytest, seed) {
if (!missing(seed)) {set.seed(seed)}
if (missing(x) && missing(y)) {
if (Ntotal<=2000) {x <- lhs::maximinLHS(Ntotal, d)}
else {x <- decentLHS(Ntotal, d, max.time=15)}
y <- apply(x, 1, f)
}
fit.time.start <- Sys.time()
mod <- MRFA::MRFA_fit(X=x, Y=y, verbose=FALSE)
fit.time.end <- Sys.time()
pred.time.start <- Sys.time()
pred <- predict(mod, xtest, lambda = min(mod$lambda))
pred.time.end <- Sys.time()
list(mean=pred$y_hat, var=rep(NaN, nrow(xtest)), n=nrow(x),
pred.time=as.numeric(pred.time.end - pred.time.start, units="secs"),
fit.time =as.numeric(fit.time.end - fit.time.start , units="secs"))
}
run_svm <- function(Ntotal, Nappend, f, d, x, y, xtest, ytest, seed) {#browser()
if (!missing(seed)) {set.seed(seed)}
if (missing(x) && missing(y)) {
if (Ntotal<=2000) {x <- lhs::maximinLHS(Ntotal, d)}
else {x <- decentLHS(Ntotal, d, max.time=15)}
y <- apply(x, 1, f)
}
require(e1071)
fit.time.start <- Sys.time()
mod <- svm(x, y)
fit.time.end <- Sys.time()
pred.time.start <- Sys.time()
pred <- predict(mod, xtest)
pred.time.end <- Sys.time()
list(mean=pred, var=rep(NaN, nrow(xtest)), n=nrow(x),
pred.time=as.numeric(pred.time.end - pred.time.start, units="secs"),
fit.time =as.numeric(fit.time.end - fit.time.start , units="secs"))
}
run_mlegp <- function(Ntotal, Nappend, f, d, x, y, xtest, ytest, seed) {#browser()
if (!missing(seed)) {set.seed(seed)}
if (missing(x) && missing(y)) {
if (Ntotal<=2000) {x <- lhs::maximinLHS(Ntotal, d)}
else {x <- decentLHS(Ntotal, d, max.time=15)}
y <- apply(x, 1, f)
}
require(mlegp)
fit.time.start <- Sys.time()
mod <- mlegp(x, y)
fit.time.end <- Sys.time()
pred.time.start <- Sys.time()
pred <- predict(mod, xtest, se.fit=T)
pred.time.end <- Sys.time()
list(mean=pred$fit, var=pred$se.fit^2, n=nrow(x),
pred.time=as.numeric(pred.time.end - pred.time.start, units="secs"),
fit.time =as.numeric(fit.time.end - fit.time.start , units="secs"))
}
run_GPfit <- function(Ntotal, Nappend, f, d, x, y, xtest, ytest, seed) {#browser()
if (!missing(seed)) {set.seed(seed)}
if (missing(x) && missing(y)) {
if (Ntotal<=2000) {x <- lhs::maximinLHS(Ntotal, d)}
else {x <- decentLHS(Ntotal, d, max.time=15)}
y <- apply(x, 1, f)
}
require(GPfit)
fit.time.start <- Sys.time()
mod <- GP_fit(x, y)
fit.time.end <- Sys.time()
pred.time.start <- Sys.time()
pred <- predict(mod, xtest)
pred.time.end <- Sys.time()
list(mean=pred$Y_hat, var=pred$MSE, n=nrow(x),
pred.time=as.numeric(pred.time.end - pred.time.start, units="secs"),
fit.time =as.numeric(fit.time.end - fit.time.start , units="secs"))
}
run_CGGP <- function(Ntotal, Nappend, Nlhs, f, d, x, y, xtest, ytest, seed, selection.method, correlation) {#browser()
require("CGGP")
if (!missing(seed)) {set.seed(seed)}
if (!missing(Nlhs) && Nlhs!=0) {stop("Nlhs given to run_CGGP")}
fit.time.start <- Sys.time()
sg <- CGGPcreate(d=d, batchsize=min(Ntotal,200), corr=correlation)
sg <- CGGPfit(sg, apply(sg$design, 1, f))
notdone <- (nrow(sg$design) < Ntotal)
while (notdone) {
# print(sg)
Nalready <- nrow(sg$design)
ni <- if (Nalready<1000) {200} else if (Nalready<10000) {500} else if (Nalready<20000) {2000} else {10000}
if (ni +Nalready > Ntotal) {
ni <- Ntotal - Nalready
notdone <- FALSE
}
sg <- CGGPappend(sg, batchsize = ni, selectionmethod = selection.method)
if (!is.null(sg$design_unevaluated)) {
ynew <- apply(sg$design_unevaluated, 1, f)
sg <- CGGPfit(sg, Ynew=ynew)
} else {
print('Nothing new to evaluate, hopefully !notdone')
}
}
fit.time.end <- Sys.time()
# print(sg)
Nevaluated <- if (!is.null(sg$design)) nrow(sg$design) else {0} + nrow(sg$Xs)
if (Nevaluated > Ntotal) {stop("CGGP has more points than Ntotal")}
pred.time.start <- Sys.time()
pred <- predict(sg, xtest)
pred.time.end <- Sys.time()
list(mean=pred$mean, var=pred$var,
n=Nevaluated,
pred.time=as.numeric(pred.time.end - pred.time.start, units="secs"),
fit.time =as.numeric(fit.time.end - fit.time.start , units="secs"))
}
run_CGGPsupp <- function(Ntotal, Nappend, Nlhs, f, d, x, y, xtest, ytest, seed, HandlingSuppData, selection.method, correlation) {#browser()
require("CGGP")
if (!missing(seed)) {set.seed(seed)}
xsup <- lhs::maximinLHS(Nlhs, d)
ysup <- apply(xsup, 1, f)
fit.time.start <- Sys.time()
sg <- CGGPcreate(d=d, batchsize=0, Xs=xsup, Ys=ysup, corr=correlation)
notdone <- (Nlhs < Ntotal)
while (notdone) {
# print(sg)
Nalready <- (if (!is.null(sg$design)) {nrow(sg$design)} else {0}) + nrow(sg$Xs)
ni <- if (Nalready<1000) {200} else if (Nalready<10000) {500} else if (Nalready<20000) {2000} else {10000}
if (ni +Nalready > Ntotal) {
ni <- Ntotal - Nalready
notdone <- FALSE
}
sg <- CGGPappend(sg, batchsize = ni, selection.method)
if (!is.null(sg$design_unevaluated)) {
ynew <- apply(sg$design_unevaluated, 1, f)
sg <- CGGPfit(sg, Ynew=ynew, Xs=xsup, Ys=ysup, HandlingSuppData = HandlingSuppData)
} else {
print('Nothing new to evaluate, hopefully !notdone')
}
}
fit.time.end <- Sys.time()
# print(sg)
Nevaluated <- if (!is.null(sg$design)) nrow(sg$design) else {0} + nrow(sg$Xs)
if (Nevaluated > Ntotal) {stop("CGGP has more points than Ntotal")}
pred.time.start <- Sys.time()
pred <- predict(sg, xtest)
pred.time.end <- Sys.time()
list(mean=pred$mean, var=pred$var,
n=Nevaluated,
pred.time=as.numeric(pred.time.end - pred.time.start, units="secs"),
fit.time =as.numeric(fit.time.end - fit.time.start , units="secs"))
}
run_CGGPsupponly <- function(Ntotal, Nappend, Nlhs, f, d, x, y, xtest, ytest, seed, correlation) {#browser()
require("CGGP")
if (!missing(seed)) {set.seed(seed)}
if (Ntotal > 2000) {stop("CGGPsupponly can't run with more than 2000")}
xsup <- lhs::maximinLHS(Ntotal, d)
ysup <- apply(xsup, 1, f)
fit.time.start <- Sys.time()
sg <- CGGPcreate(d=d, batchsize=0, Xs=xsup, Ys=ysup, corr=correlation)
fit.time.end <- Sys.time()
# print(sg)
Nevaluated <- nrow(sg$Xs)
if (Nevaluated > Ntotal) {stop("CGGP has more points than Ntotal")}
pred.time.start <- Sys.time()
pred <- predict(sg, xtest)
pred.time.end <- Sys.time()
list(mean=pred$mean, var=pred$var,
n=Nevaluated,
pred.time=as.numeric(pred.time.end - pred.time.start, units="secs"),
fit.time =as.numeric(fit.time.end - fit.time.start , units="secs"))
}
run_CGGPoneshot <- function(Ntotal, Nappend, Nlhs, f, d, x, y, xtest, ytest, seed, correlation) {#browser()
require("CGGP")
if (!missing(seed)) {set.seed(seed)}
# xsup <- lhs::maximinLHS(Nlhs, d)
# ysup <- apply(xsup, 1, f)
fit.time.start <- Sys.time()
sg <- CGGPcreate(d=d, batchsize=Ntotal, corr=correlation)
sg <- CGGPfit(sg, Y=apply(sg$design, 1, f))
fit.time.end <- Sys.time()
# print(sg)
Nevaluated <- nrow(sg$design)
if (Nevaluated > Ntotal) {stop("CGGP has more points than Ntotal")}
pred.time.start <- Sys.time()
pred <- predict(sg, xtest)
pred.time.end <- Sys.time()
list(mean=pred$mean, var=pred$var,
n=Nevaluated,
pred.time=as.numeric(pred.time.end - pred.time.start, units="secs"),
fit.time =as.numeric(fit.time.end - fit.time.start , units="secs"))
}
# Need a generic function that passes to specific ones
run_one <- function(package, selection.method, correlation, HandlingSuppData,
f, d, npd, replicate) {#browser()
# package <- psch$package
n <- npd * d
# if (n!= 500) {stop('bad n')}
f <- eval(parse(text=paste0("TestFunctions::", f)))
ntest <- 1e4
# xtest <- matrix(runif(ntest*d), ncol=d)
xtest <- decentLHS(ntest, d, max.time = 10)
ytest <- apply(xtest, 1, f)
# if (package == "CGGP") {
# out <- run_CGGP(Nappend=floor(n * (1:5)/5), f=f, d=d, xtest=xtest)
if (package == "CGGPsupp") {
# out <- run_CGGP(Nappend=floor(n*(2:5)/5), Nlhs=floor(.2*n), f=f, d=d, xtest=xtest)
out <- run_CGGPsupp(Ntotal=n, Nlhs=10*d, f=f, d=d, xtest=xtest, HandlingSuppData=HandlingSuppData, selection.method=selection.method, correlation=correlation)
# } else if (package == "CGGPsupponly") {
# out <- run_CGGP(Nappend=c(), Nlhs=n, f=f, d=d, xtest=xtest)
} else if (package == "CGGPsupponly") {
out <- run_CGGPsupponly(Ntotal=n, f=f, d=d, xtest=xtest, correlation=correlation)
} else if (package == "CGGPoneshot") {
out <- run_CGGPoneshot(Ntotal=n, f=f, d=d, xtest=xtest, correlation=correlation)
} else if (package == "CGGP") {
out <- run_CGGP(Ntotal=n, f=f, d=d, xtest=xtest, selection.method=selection.method, correlation=correlation)
} else if (package == "laGP") {
out <- run_lagp(Ntotal=n, f=f, d=d, xtest=xtest)
} else if (package == "aGP") {
out <- run_lagp_bobby(Ntotal=n, f=f, d=d, xtest=xtest, use_agp=TRUE)
} else if (package == "MRFA") {
out <- run_MRFA(Ntotal=n, f=f, d=d, xtest=xtest)
} else if (package == "svm") {
out <- run_svm(Ntotal=n, f=f, d=d, xtest=xtest)
} else if (package == "mlegp") {
out <- run_mlegp(Ntotal=n, f=f, d=d, xtest=xtest)
} else if (package == "GPfit") {
out <- run_GPfit(Ntotal=n, f=f, d=d, xtest=xtest)
} else {
stop(paste("Package", package, "not recognized"))
}
# browser()
outstats <- CGGP::valstats(predmean=out[[1]], predvar=out[[2]],Yval=ytest) #, bydim=FALSE)
if (out$n > n) {warning(paste("n too big for", package, n, f, d))}
outstats
}
expand.grid.df <- function(...) Reduce(function(...) merge(..., by=NULL), list(...))
eg1 <- expand.grid(selection.method = c("UCB", "Greedy"),
correlation = c("CauchySQ", "Matern32", "PowerExp"), stringsAsFactors=F)
eg2a <- expand.grid.df(eg1, data.frame(HandlingSuppData=c("Ignore", "Correct"), stringsAsFactors=F), data.frame(package="CGGPsupp", stringsAsFactors=F))
eg2b <- expand.grid.df(eg1, data.frame(HandlingSuppData="NA", stringsAsFactors=F), data.frame(package=c("CGGP"), stringsAsFactors=F))
eg2c <- expand.grid(selection.method="NA", correlation = c("CauchySQ", "Matern32", "PowerExp"), HandlingSuppData="NA", package=c("CGGPoneshot", "CGGPsupponly"), stringsAsFactors=F)
eg2d <- data.frame(selection.method="NA", correlation="NA", HandlingSuppData="NA",
package=c("MRFA", "svm", "aGP", "laGP", "mlegp", "GPfit"), stringsAsFactors=F)
eg3 <- rbind(eg2a, eg2b, eg2c, eg2d)
require("comparer")
excomp <- ffexp$new(
eval_func = run_one,
varlist = c("decentLHS", "run_CGGP", "run_CGGPoneshot", "run_CGGPsupp",
"run_CGGPsupponly", "run_GPfit", "run_lagp", "run_lagp_bobby",
"run_mlegp", "run_MRFA", "run_svm"),
fd=data.frame(f=c("beambending","OTL_Circuit","piston","borehole","wingweight"),
d=c(3,6,7,8,10),
row.names = c("beam","OTL","piston","borehole","wingweight"), stringsAsFactors = F),
# package=c("CGGPsupp", "CGPPoneshot", "CGGPsupponly", "CGGP",
# "MRFA", "svm", "aGP", "laGP", "mlegp", "GPfit"),
psch=eg3,
npd=c(10, 30, 100, 300, 1000, 3000, 10000),
parallel=TRUE,
parallel_cores = 20,
replicate=1:10, #:5,
folder_path= "/home/collin/scratch/SGGP/scratch/ExternalComparison/ExComp4"
# folder_path="./scratch/ExternalComparison/ExComp4/"
)
# Remove ones that can't do full size
package.name <- excomp$arglist$psch$package[excomp$rungrid$psch]
npd.excomp <- excomp$arglist$npd[excomp$rungrid$npd]
n.excomp <- npd.excomp * excomp$arglist$fd$d[excomp$rungrid$fd]
excomp$completed_runs[package.name == "CGGPsupponly" & n.excomp > 1000] <- TRUE
excomp$completed_runs[package.name == "laGP" & n.excomp > 1000] <- TRUE
excomp$completed_runs[package.name == "mlegp" & n.excomp > 400] <- TRUE
excomp$completed_runs[package.name == "GPfit" & n.excomp > 100] <- TRUE
# agp is giving errors
excomp$completed_runs[package.name == "aGP"] <- TRUE
table(paste(package.name, excomp$completed_runs))
# plyr::ddply(data.frame(package.name, compruns=excomp$completed_runs), "package.name", function(d) {data.frame(done=sum(d$compruns), notdone=sum(!d$compruns))})
rbind(plyr::ddply(data.frame(package.name, compruns=excomp$completed_runs), "package.name", function(d) {data.frame(done=sum(d$compruns), notdone=sum(!d$compruns))}), data.frame(package.name="all", done=sum(excomp$completed_runs), notdone=sum(!excomp$completed_runs)))
# excomp$run_one(1752)
# excomp$run_one(7299)
# excomp$run_all(save_output = T, parallel = F, parallel_temp_save = T, run_order = "random")
# excomp$rungrid
# try because it gave delete error before, but shouldn't need it now
table(excomp$completed_runs)
try(excomp$recover_parallel_temp_save(delete_after = FALSE))
table(excomp$completed_runs)
# plyr::ddply(data.frame(package.name, compruns=excomp$completed_runs), "package.name", function(d) {data.frame(done=sum(d$compruns), notdone=sum(!d$compruns))})
rbind(plyr::ddply(data.frame(package.name, compruns=excomp$completed_runs), "package.name", function(d) {data.frame(done=sum(d$compruns), notdone=sum(!d$compruns))}), data.frame(package.name="all", done=sum(excomp$completed_runs), notdone=sum(!excomp$completed_runs)))
excomp$save_self()
# excomp$run_all(parallel_temp_save = TRUE, delete_parallel_temp_save_after=FALSE,
# write_start_files=T, write_error_files=T)
# Getting errors, run by package.name to see which is causing it
excomp$parallel_cores <- 10
excomp$run_all(to_run = which(!excomp$completed_runs & (package.name == "mlegp")),
parallel_temp_save = TRUE, delete_parallel_temp_save_after=FALSE,
write_start_files=T, write_error_files=T)
# excomp$run_all()
cat("Completed all runs in ExternalComparer4.R\n")
excomp$save_self()
cat("Saved self\n")
if (F) {
excomp <- readRDS("./scratch/ExternalComparison/ExternalComparer4_mostlydone.rds")
excompbass <- readRDS("./scratch/ExternalComparison/ExternalComparer4bass_quarterdone.rds")
excomp$plot_run_times()
plyr::dlply(excomp$outcleandf, "d")
require('ggplot2')
ecdf <- rbind(excomp$outcleandf[excomp$completed_runs & !is.na(excomp$outcleandf$package),],
excompbass$outcleandf[excompbass$completed_runs & !is.na(excompbass$outcleandf$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()
# saveRDS(excomp, "./scratch/ExternalComparison/ExComp1_completed.rds")
}
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.