library(testthat) Sys.setenv('OMP_THREAD_LIMIT'=2) library(rlibkriging) ##library(rlibkriging, lib.loc="bindings/R/Rlibs") ##library(testthat) context ("no noise/nugget") f <- function(x) { 1 - 1 / 2 * (sin(12 * x) / (1 + x) + 2 * cos(7 * x) * x^5 + 0.7) } plot(f) n <- 5 X_o <- seq(from = 0, to = 1, length.out = n) noise = 0 sigma2 = 0.01 set.seed(1234) y_o <- f(X_o) + rnorm(n, sd = sqrt(noise)) points(X_o, y_o,pch=16) lk_no <- NoiseKriging(y = matrix(y_o, ncol = 1), noise = matrix(rep(noise, n), ncol = 1), X = matrix(X_o, ncol = 1), kernel = "gauss", regmodel = "linear", optim = "none", normalize = FALSE, parameters = list(theta = matrix(0.1), sigma2 = sigma2)) lk_nu <- NuggetKriging(y = matrix(y_o, ncol = 1), X = matrix(X_o, ncol = 1), kernel = "gauss", regmodel = "linear", optim = "none", normalize = FALSE, parameters = list(theta = matrix(0.1), nugget=noise, sigma2=sigma2)) lk = Kriging(y = matrix(y_o, ncol = 1), X = matrix(X_o, ncol = 1), kernel = "gauss", regmodel = "linear", optim = "none", normalize = FALSE, parameters = list(theta = matrix(0.1), sigma2 = sigma2)) test_that("Consistency between Noise(0)Kriging and Kriging", { expect_equal(lk$T(), lk_no$T()/sqrt(sigma2)) expect_equal(lk$M(), lk_no$M()*sqrt(sigma2)) expect_equal(lk$beta(), lk_no$beta()) expect_equal(lk$z(), lk_no$z()*sqrt(sigma2)) }) test_that("Consistency between Nugget(0)Kriging and Kriging", { expect_equal(lk$T(), lk_nu$T()) expect_equal(lk$M(), lk_nu$M()) expect_equal(lk$beta(), lk_nu$beta()) expect_equal(lk$z(), lk_nu$z()) }) ## Simulate X_n = seq(0,1,,5) ls = lk$simulate(100, 123, X_n, will_update=FALSE) #y_u = rs[i_u,1] # force matching 1st sim ls_no = lk_no$simulate(100, 123, X_n, with_noise=NULL, will_update=FALSE) ls_nu = lk_nu$simulate(100, 123, X_n, with_nugget=FALSE, will_update=FALSE) plot(f) points(X_o,y_o,pch=16) for (i in 1:length(X_o)) { lines(c(X_o[i],X_o[i]),c(y_o[i]+2*sqrt(noise),y_o[i]-2*sqrt(noise)),col='black',lwd=4) } for (i in 1:min(10,ncol(ls))) { lines(X_n,ls[,i],col=rgb(0,0,0,.51),lwd=2) lines(X_n,ls_nu[,i],col=rgb(1,0.5,0,.51),lwd=4,lty=2) lines(X_n,ls_no[,i],col=rgb(1,0,0.5,.51),lwd=6,lty=2) } for (i in 1:1:length(X_n)) { ds=density(ls[i,]) ds_nu=density(ls_nu[i,]) ds_no=density(ls_no[i,]) polygon( X_n[i] + ds$y/20, ds$x, col=rgb(0,0,0,0.2),border=NA) polygon( X_n[i] + ds_nu$y/20, ds_nu$x, col=rgb(0,0,1,0.2),border=NA) polygon( X_n[i] + ds_no$y/20, ds_no$x, col=rgb(1,0,0,0.2),border=NA) #test_that(desc="updated,simulated sample follows simulated,updated distribution", # expect_true(ks.test(lus[i,],lsu[i,])$p.value > 0.01)) } for (i in 1:length(X_n)) { plot(density(ls[i,]),xlim=range(ls[i,], c(ls_no[i,],ls_nu[i,]))) lines(density(ls_no[i,]),col='orange') lines(density(ls_nu[i,]),col='red') if (sd(ls_no[i,])>1e-3 && sd(ls_nu[i,])>1e-3) # otherwise means that density is ~ dirac, so don't test test_that(desc=paste0("updated,simulated sample follows simulated,updated distribution ", mean(ls_no[i,]),",",sd(ls_no[i,])," != ",mean(ls_nu[i,]),",",sd(ls_nu[i,])), expect_gt(suppressWarnings(ks.test(ls_no[i,],ls_nu[i,])$p.value), 0.01)) # just check that it is not clearly wrong } ## Update simulate #i_u = c(9,13) #X_u = X_n[i_u]# c(.4,.6) X_u = c(.4,.6) y_u = f(X_u) + rnorm(length(X_u), sd = sqrt(noise)) noise_u = rep(noise, length(X_u)) ls = lk$simulate(100, 123, X_n, will_update=TRUE) #y_u = rs[i_u,1] # force matching 1st sim lus=NULL lus = lk$update_simulate(y_u, X_u) ls_no = lk_no$simulate(100, 123, X_n, will_update=TRUE) lus_no=NULL lus_no = lk_no$update_simulate(y_u, noise_u, X_u) ls_nu = lk_nu$simulate(100, 123, X_n, will_update=TRUE) lus_nu=NULL lus_nu = lk_nu$update_simulate(y_u, X_u) lu = copy(lk) lu$update(y_u, matrix(X_u,ncol=1), refit=FALSE) lsu=NULL lsu = lu$simulate(100, 123, X_n) plot(f) points(X_o,y_o,pch=16) for (i in 1:length(X_o)) { lines(c(X_o[i],X_o[i]),c(y_o[i]+2*sqrt(noise),y_o[i]-2*sqrt(noise)),col='black',lwd=4) } points(X_u,y_u,col='red',pch=16) for (i in 1:length(X_u)) { lines(c(X_u[i],X_u[i]),c(y_u[i]+2*sqrt(noise),y_u[i]-2*sqrt(noise)),col='red',lwd=4) } for (i in 1:min(10,ncol(lus))) { lines(X_n,ls[,i],col=rgb(0,0,0,.1),lwd=2) lines(X_n,lus_nu[,i],col=rgb(1,0.5,0,.51),lwd=4,lty=2) lines(X_n,lus_no[,i],col=rgb(1,0,0.5,.51),lwd=6,lty=2) lines(X_n,lus[,i],col=rgb(1,0,0,.51),lwd=2) lines(X_n,lsu[,i],col=rgb(0,0,1,.51),lwd=2) } for (i in 1:length(X_n)) { plot(density(lus[i,]),xlim=range(lus[i,], c(lus_no[i,],lus_nu[i,]))) lines(density(lus_no[i,]),col='orange') lines(density(lus_nu[i,]),col='red') if (sd(lus_no[i,])>1e-3 && sd(lus_nu[i,])>1e-3) # otherwise means that density is ~ dirac, so don't test test_that(desc=paste0("updated,simulated sample follows simulated,updated distribution ", mean(lus_no[i,]),",",sd(lus_no[i,])," != ",mean(lus_nu[i,]),",",sd(lus_nu[i,])), expect_true(ks.test(lus_no[i,],lus_nu[i,])$p.value > 0.001)) # just check that it is not clearly wrong } ########################################## noise/nugget >0 ########################################## context ("noise = nugget >0") f <- function(x) { 1 - 1 / 2 * (sin(12 * x) / (1 + x) + 2 * cos(7 * x) * x^5 + 0.7) } plot(f) n <- 5 X_o <- seq(from = 0, to = 1, length.out = n) noise = 0.01 sigma2 = 0.09 set.seed(1234) y_o <- f(X_o) + rnorm(n, sd = sqrt(noise)) points(X_o, y_o,pch=16) lk_no <- NoiseKriging(y = matrix(y_o, ncol = 1), noise = matrix(rep(noise, n), ncol = 1), X = matrix(X_o, ncol = 1), kernel = "gauss", regmodel = "linear", optim = "none", #normalize = TRUE, parameters = list(theta = matrix(0.1), sigma2 = sigma2)) lk_nu <- NuggetKriging(y = matrix(y_o, ncol = 1), X = matrix(X_o, ncol = 1), kernel = "gauss", regmodel = "linear", optim = "none", #normalize = TRUE, parameters = list(theta = matrix(0.1), nugget=noise, sigma2=sigma2)) test_that("Consistency between Noise(0)Kriging and Kriging", { expect_equal(lk_nu$T(), lk_no$T()/sqrt(noise+sigma2)) expect_equal(lk_nu$M(), lk_no$M()*sqrt(noise+sigma2)) expect_equal(lk_nu$beta(), lk_no$beta()) expect_equal(lk_nu$z(), lk_no$z()*sqrt(noise+sigma2)) }) ## Simulate X_n = seq(0,1,,11) ls_no = NULL ls_no = lk_no$simulate(10, 123, X_n, with_noise = 0, will_update=FALSE) ls_nu = NULL ls_nu = lk_nu$simulate(10, 123, X_n, with_nugget=FALSE, will_update=FALSE) plot(f,xlim=c(-0.1,0.1)) points(X_o,y_o,pch=16) for (i in 1:length(X_o)) { lines(c(X_o[i],X_o[i]),c(y_o[i]+2*sqrt(noise),y_o[i]-2*sqrt(noise)),col='black',lwd=4) } for (i in 1:min(100,ncol(ls_nu))) { lines(X_n,ls_nu[,i],col='red',lwd=3,lty=2) lines(X_n,ls_no[,i],col='blue',lty=2) } for (i in 1:1:length(X_n)) { ds_nu=density(ls_nu[i,]) ds_no=density(ls_no[i,]) polygon( X_n[i] + ds_nu$y/20, ds_nu$x, col=rgb(0,0,1,0.2),border=NA) polygon( X_n[i] + ds_no$y/20, ds_no$x, col=rgb(1,0,0,0.2),border=NA) #test_that(desc="updated,simulated sample follows simulated,updated distribution", # expect_true(ks.test(lus[i,],lsu[i,])$p.value > 0.01)) } for (i in 1:length(X_n)) { plot(density(ls_no[i,]),xlim=range(c(ls_no[i,],ls_nu[i,]))) lines(density(ls_no[i,]),col='orange') lines(density(ls_nu[i,]),col='red') if (sd(ls_no[i,])>1e-3 && sd(ls_nu[i,])>1e-3) # otherwise means that density is ~ dirac, so don't test test_that(desc=paste0("nugget & noise simulated are consistent: ", mean(ls_no[i,]),",",sd(ls_no[i,])," != ",mean(ls_nu[i,]),",",sd(ls_nu[i,])), expect_true(ks.test(ls_no[i,],ls_nu[i,])$p.value > 0.001)) # just check that it is not clearly wrong } ## Update simulate #i_u = c(9,13) #X_u = X_n[i_u]# c(.4,.6) X_u = c(.4,.6) y_u = f(X_u) + rnorm(length(X_u), sd = sqrt(noise)) noise_u = rep(noise, length(X_u)) ls_no = lk_no$simulate(100, 123, X_n, will_update=TRUE) lus_no=NULL lus_no = lk_no$update_simulate(y_u, noise_u, X_u) ls_nu = lk_nu$simulate(100, 123, X_n, with_nugget=FALSE, will_update=TRUE) lus_nu=NULL lus_nu = lk_nu$update_simulate(y_u, X_u) plot(f,xlim=c(-0.1,0.1)) points(X_o,y_o,pch=16) for (i in 1:length(X_o)) { lines(c(X_o[i],X_o[i]),c(y_o[i]+2*sqrt(noise),y_o[i]-2*sqrt(noise)),col='black',lwd=4) } points(X_u,y_u,col='red',pch=16) for (i in 1:length(X_u)) { lines(c(X_u[i],X_u[i]),c(y_u[i]+2*sqrt(noise),y_u[i]-2*sqrt(noise)),col='red',lwd=4) } for (i in 1:min(10,ncol(lus_nu))) { lines(X_n,lus_nu[,i],col=rgb(1,0.5,0,.51),lwd=4,lty=2) lines(X_n,lus_no[,i],col=rgb(1,0,0.5,.51),lwd=4,lty=2) } for (i in 1:1:length(X_n)) { dus_nu=density(lus_nu[i,]) dus_no=density(lus_no[i,]) polygon( X_n[i] + dus_nu$y/100, dus_nu$x, col=rgb(0,0,1,0.2),border=NA) polygon( X_n[i] + dus_no$y/100, dus_no$x, col=rgb(1,0,0,0.2),border=NA) #test_that(desc="updated,simulated sample follows simulated,updated distribution", # expect_true(ks.test(lus[i,],lsu[i,])$p.value > 0.01)) } for (i in 1:length(X_n)) { plot(density(ls_nu[i,]),xlim=range(c(ls_no[i,],ls_nu[i,],lus_no[i,],lus_nu[i,]))) lines(density(ls_no[i,]),col='blue') lines(density(lus_no[i,]),col='orange') lines(density(lus_nu[i,]),col='red') #if (all(abs(X_n[i]-X_o)>0)) # means we are not on design points (where nuggetK is deterministic, while noiseK is not) if (sd(lus_no[i,])>1e-3 && sd(lus_nu[i,])>1e-3) # otherwise means that density is ~ dirac, so don't test test_that(desc=paste0("nugget & noise updated simulate are consistent at x=",X_n[i]," ", mean(lus_no[i,]),",",sd(lus_no[i,])," != ",mean(lus_nu[i,]),",",sd(lus_nu[i,])), expect_true(ks.test(lus_no[i,],lus_nu[i,])$p.value > 0.000001)) # just check that it is not clearly wrong }