{"id":1961,"date":"2019-09-09T00:51:30","date_gmt":"2019-09-08T15:51:30","guid":{"rendered":"http:\/\/141.164.34.82\/?p=1961"},"modified":"2019-09-09T01:06:43","modified_gmt":"2019-09-08T16:06:43","slug":"black-box-%eb%aa%a8%ed%98%95-%ec%82%b4%ed%8e%b4%eb%b3%b4%ea%b8%b0-%eb%aa%a8%ed%98%95%ec%9d%98-%ec%98%88%ec%b8%a1-%ec%9e%94%ec%b0%a8-02","status":"publish","type":"post","link":"http:\/\/ds.sumeun.org\/?p=1961","title":{"rendered":"BLACK BOX \ubaa8\ud615 \uc0b4\ud3b4\ubcf4\uae30: \ubaa8\ud615\uc758 \uc608\uce21, \uc794\ucc28 02"},"content":{"rendered":"
library(dplyr)\nlibrary(ggplot2)\nlibrary(randomForest)\n\nTURE <- TRUE\n<\/code><\/pre>\n
\ubaa8\ud615, \uc794\ucc28<\/h2>\n
\uc5ec\uae30\uc11c\ub294 ICEbox<\/code> \ud328\ud0a4\uc9c0\uc640 plotmo<\/code> \ud328\ud0a4\uc9c0\ub97c \uc0ac\uc6a9\ud558\uc5ec \ubaa8\ud615\uc744 \uc880 \ub354 \uc0b4\ud3b4\ubcf8\ub2e4.<\/p>\n
\uc608\uc81c \ubaa8\ud615<\/h2>\n
\uc5ec\uae30\uc11c\ub294 \uc9c0\ub09c \ubc88<\/a>, \uadf8\ub9ac\uace0 \uc9c0\uc9c0\ub09c\ubc88<\/a>\uc758 \uc608\uc81c \ubaa8\ud615\uc744 \uadf8\ub300\ub85c \uc0ac\uc6a9\ud55c\ub2e4. <\/p>\n
\ub370\uc774\ud130 \uc0dd\uc131 \ubaa8\ud615\uc740 \ub2e4\uc74c\uacfc \uac19\ub2e4.<\/p>\n
\\[y = x_1^2 + \\frac{x_2}{|x_1|+0.1} + \\Big(I(x_2<0)+I(x_2<1)\\times x_3\\Big) \\times \\frac{1}{5}(x_4+x_5+x_6+x_7+x_8)+e\\]<\/p>\n
\\[e\\sim\\mathcal{N}(0,1)\\]<\/p>\n
\uc774\ub54c \\(x_3\\) \uc740 \\(-1\\), \\(+1\\) \uac12\ub9cc \uac00\uc9c8 \uc218 \uc788\uace0, \ub098\uba38\uc9c0 \ubcc0\uc218\ub4e4\uc740 \ubaa8\ub450 \uc5f0\uc18d\ud615\uc774\ub2e4.<\/p>\n
genData = function(N=1000, iseed=100) {\n  set.seed(iseed)\n  sdErr = 1\n\n  x1 <- runif(N, -3, 3)\n  x11 <- x1^2 \n\n  x2 <- runif(N, -3, 3)\n  x3 <- sample(c(-1, 1), N, replace=TRUE)\n\n  ncolMatX <- 5\n  ncolMatX2 <- 2\n\n  require(mvtnorm)\n  sigma <- matrix(\n    rep(0.97, ncolMatX*ncolMatX),\n    ncolMatX, ncolMatX)\n  diag(sigma) = 1\n\n  matX <- rmvnorm(N, mean=rep(0,ncolMatX), sigma=sigma)\n  x40 <- apply(matX, 1, mean)\/ncolMatX\n  colnames(matX) <- paste0('x', 4:(ncolMatX+3))\n\n  matX2 <- rmvnorm(N, mean=rep(0,ncolMatX2), sigma=diag(ncolMatX2))\n  colnames(matX2) <- paste0('x', (ncolMatX+4):(ncolMatX+ncolMatX2+3))\n\n  y <- x11 + x2\/(abs(x1)+0.1) + (x2 < 0)*x40 + (x2<1)*x3 * x40 + rnorm(N, 0, sdErr)\n\n  return(data.frame(y, x1, x2, x3, matX, matX2))\n}\n\ndat <- genData()\n<\/code><\/pre>\n
\uc608\uc81c \ubaa8\ud615\uc758 \uc2e4\ub840\uc778 dat<\/code>\uc5d0\ub294 \uacb0\uacfc \ubcc0\uc218 y<\/code>\uc640 \uc124\uba85\ubcc0\uc218 (\ud6c4\ubcf4) x1<\/code> ~ x10<\/code>\uc774 \uc800\uc7a5\ub418\uc5b4 \uc788\ub2e4. \uacb0\uacfc \ubcc0\uc218 y<\/code>\uc640 \uc124\uba85\ubcc0\uc218 x1<\/code> ~ x10<\/code>\uc758 \uad00\uacc4\ub294 \ub2e4\uc74c\uacfc \uac19\ub2e4.<\/p>\n
pairs(dat)\n<\/code><\/pre>\n
\"plot<\/p>\n
pairs(dat[,c("y", "x1", "x2", "x3")])\n<\/code><\/pre>\n
\"plot<\/p>\n
pairs(dat[,c("y", paste0("x", 4:10))])\n<\/code><\/pre>\n
\"plot<\/p>\n
\uc704\uc758 \uc0b0\uc810\ub3c4\ub97c \ubcf4\uba74 \uc54c\uaca0\uc9c0\ub9cc, x1<\/code>\uacfc x2<\/code>\ub97c \uc81c\uc678\ud558\uace0\uc11c\ub294 y<\/code>\uc640 \uc124\uba85\ubcc0\uc218\uc758 \uad00\uacc4\ub97c \ub69c\ub837\uc774 \uc54c\uc544\ub0b4\uae30\ub294 \ud798\ub4e4\ub2e4. \uadf8\ub9ac\uace0 x1<\/code>\uacfc y<\/code>\uc758 \uad00\uacc4\ub3c4 \uc0ac\uc2e4 \uc5b4\ub5a4 \uad00\uacc4\ub77c\uace0 \ub69c\ub837\uc774 \uc124\uba85\ud558\uae30 \ud798\ub4e4\ub2e4. \uc0b0\uc810\ub3c4\uc640 \ud68c\uadc0\uc120\uc744 \uadf8\ub824\ubcf4\uba74 \ub2e4\uc74c\uacfc \uac19\ub2e4. <\/p>\n
dat %>% ggplot(aes(x=x1, y=y)) + geom_point() + geom_smooth(method='loess', span=0.3)\n<\/code><\/pre>\n
\"plot<\/p>\n
\uc55e\uc5d0\uc11c \ud68c\uadc0\ubd84\uc11d\uc744 \ud55c \ubc88 \ub3cc\ub824\ubcf4\uac70\ub098 \ub79c\ub364\ud3ec\ub808\uc2a4\ud2b8\ub85c \ubcc0\uc218 \uc911\uc694\ub3c4\ub97c \uc54c\uc544\ubcf4\uace0<\/a>, \ubaa8\ud615\uacfc \uc794\ucc28\ub97c \uc880 \ub354 \uc798 \uc774\ud574\ud558\ub294 \ubc29\ubc95\uc744 \uc2dc\uc5f0\ud574\ubcf4\uc558\ub2e4<\/a>.<\/p>\n
\ud2b9\ud788 \uc9c0\ub09c\ubc88\uc5d0 \uc124\uba85\ud55c \ubaa8\ud615\uacfc \uc794\ucc28 \uc2dc\uac01\ud654 \ubc29\ubc95\uc740 \ub79c\ub364\ud3ec\ub808\uc2a4\ud2b8 \ubfd0 \uc544\ub2c8\ub77c \uadf8\ub798\ub514\uc5b8\ud2b8 \ud2b8\ub9ac \ubd80\uc2a4\ud305\uacfc \uc778\uacf5 \uc2e0\uacbd\ub9dd \ub4f1\uc5d0\ub3c4 \uc801\uc6a9\ud560 \uc218 \uc788\ub294 \ubc94\uc6a9 \ud568\uc218\ub4e4\uc774\uc5c8\ub2e4.<\/p>\n
\uc5ec\uae30\uc11c\ub3c4 \ud754\ud788 \ub9d0\ud558\ub294 \ube14\ub799 \ubc15\uc2a4 \ubaa8\ud615\uc744 \uc2dc\uac01\ud654\ud558\uace0, \uc790\ub8cc\uc640\uc758 \uc801\ud569\uc131\uc744 \uc54c\uc544\ubcf4\ub294 \ud568\uc218\ub4e4\uc744 \uc18c\uac1c\ud55c\ub2e4. <\/p>\n
\ube44\uad50\ub97c \uc704\ud558\uc5ec \uac19\uc740 \ub370\uc774\ud130\ub97c \ub2e4\ub978 \ubaa8\ud615\uc73c\ub85c \uc801\ud569\uc2dc\ucf1c \ubcf8\ub2e4.<\/p>\n
fitLM <- lm(y ~ ., data=dat) ## linear model\n\nfitRF <- randomForest(y ~ ., data=dat) ## Random Forest Model\n\nlibrary(gam) ## gam # install.packages('gam')\nfitGAM <- gam(y ~ s(x1)+s(x2)+x3+s(x4)+s(x5)+s(x6)+s(x7)+s(x8)+s(x9)+s(x10),\n              data=dat) # x3 is a categorical variable\n\nlibrary(gbm) # install.packages('gbm')\nfitGBM <- gbm(y ~ ., data=dat, dist="gaussian", inter=2, n.trees=1000)\n\nlibrary(xgboost)\nfitXGB <- xgboost(as.matrix(dat %>% select(-y)), label=dat[,"y"], \n                  objective = "reg:linear", verbose=0,\n                  nrounds=1000)\n\nlibrary(nnet) ## nnet\nset.seed(4)\nfitNN10 <- nnet(y ~ ., data=dat,\n                linout=TRUE, size=10, decay=0.01, maxit=1000, \n                trace=FALSE)\n#fitNN20 <- nnet(y ~ ., linout=TRUE, size=20, data=dat)\nfitNN40 <- nnet(y ~ ., linout=TRUE, size=40, data=dat, maxit=1000, trace=FALSE)\n#fitNN80 <- nnet(y ~ ., linout=TRUE, size=80, data=dat)\n#fitNN160 <- nnet(y ~ ., data=dat, \n#                 linout=TRUE, size=160, \n#                 MaxNWts= 10000, trace=FALSE,\n#                 decay=0.01, maxit=10000)\n<\/code><\/pre>\n
I<\/strong>ndividual C<\/strong>onditional E<\/strong>xpectation plot toolbox<\/strong>(ICEbox<\/code>)<\/h2>\n
\uc9c0\ub09c \ud3ec\uc2a4\ud2b8\uc5d0\uc11c\ub3c4 \uc18c\uac1c\ub41c ICE<\/strong>\ub294 ICEbox<\/code> \ud328\ud0a4\uc9c0\ub97c \uc0ac\uc6a9\ud558\uc5ec \uc2dc\uac01\ud654\ud560 \uc218\ub3c4 \uc788\ub2e4. ICEbox::ice<\/code>\ub294 pdp::partial(  , ice=TRUE)<\/code>\uacfc \ub2ec\ub9ac \ub370\uc774\ud130\ub3c4 \ud568\uaed8 \uadf8\ub824\ubcf4\uc5ec \uc900\ub2e4\ub294 \uc7a5\uc810\uc774 \uc788\ub2e4. <\/p>\n
require(ICEbox)\niceRF = ice(object = fitRF, X = dat %>% select(-y), y = dat$y,\n            predictor = "x1",\n            frac_to_build = .1, verbose=FALSE)\nplot(iceRF)\n<\/code><\/pre>\n
\"plot<\/p>\n
frac_to_build=<\/code>\ub97c \ud1b5\ud574 \uc790\ub8cc\uc758 \uba87 %\ub97c \ud65c\uc6a9\ud560 \uac83\uc778\uc9c0\ub97c \uacb0\uc815\ud55c\ub2e4. <\/p>\n
require(ICEbox)\niceRF30 = ice(object = fitRF, X = dat %>% select(-y), y = dat$y,\n            predictor = "x1",\n            frac_to_build = .3, verbose=FALSE)\nplot(iceRF30)\n<\/code><\/pre>\n
\"plot<\/p>\n
\ub2e4\ub978 \ubaa8\ud615\ub4e4\uc758 \uacbd\uc6b0\ub3c4 \uc0b4\ud3b4\ubcf4\uc790. <\/p>\n
plot(iceLM <- ice(fitLM, X = dat %>% select(-y), y = dat$y, predictor = "x1", frac_to_build = .1, verbose=FALSE))\n<\/code><\/pre>\n
\"plot<\/p>\n
plot(iceRF <- ice(fitRF, X = dat %>% select(-y), y = dat$y, predictor = "x1", frac_to_build = .1, verbose=FALSE))\n<\/code><\/pre>\n
\"plot<\/p>\n
plot(iceGAM <- ice(fitGAM, X = dat %>% select(-y), y = dat$y, predictor = "x1", frac_to_build = .1, verbose=FALSE))\n<\/code><\/pre>\n
\"plot<\/p>\n
#plot(iceGBM <- ice(fitGBM, X = dat %>% select(-y), y = dat$y, predictor = "x1", frac_to_build = .1, n.trees=1000, verbose=FALSE))\n\nplot(iceXGB <- ice(fitXGB, X = as.matrix(dat %>% select(-y)), y = dat$y, predictor = "x1", frac_to_build = .1, n.trees=1000, verbose=FALSE))\n<\/code><\/pre>\n
\"plot<\/p>\n
plot(iceNN10 <- ice(fitNN10, X = dat %>% select(-y), y = dat$y, predictor = "x1", frac_to_build = .1, verbose=FALSE))\n<\/code><\/pre>\n
\"plot<\/p>\n
plot(iceNN40 <- ice(fitNN40, X = dat %>% select(-y), y = dat$y, predictor = "x1", frac_to_build = .1, verbose=FALSE))\n<\/code><\/pre>\n
\"plot<\/p>\n
clusterICE<\/code>\ub294 ICE<\/strong> \uace1\uc120\uc744 \uad70\uc9d1\ud654\ud558\uc5ec \ubcf4\uc5ec\uc900\ub2e4. \ubb34\uc218\ud788 \uc911\ucca9\ub418\uc5b4 \uc788\ub294 ICE \uace1\uc120\uc744 \uba87 \uac1c\uc758 \uad70\uc9d1\uc73c\ub85c \ub098\ub208 \ud6c4 \ub300\ud45c \uace1\uc120\uc744 \ucd94\ub824 \ubcf4\uc5ec\uc8fc\ub294 \uac83\uc774\ub2e4. \ub2e4\uc74c\uc740 \uc704\uc758 ICE \uace1\uc120\uc744 4\uac1c\uc758 \uad70\uc9d1\uc73c\ub85c \ub098\ub220\uc11c \uc2dc\uac01\ud654\ud55c\ub2e4. <\/p>\n
clusterICE(iceLM, nClusters = 4, plot_legend = TRUE, center=TRUE)\n<\/code><\/pre>\n
\"plot<\/p>\n
clusterICE(iceRF, nClusters = 4, plot_legend = TRUE, center=TRUE)\n<\/code><\/pre>\n
\"plot<\/p>\n
clusterICE(iceGAM, nClusters = 4, plot_legend = TRUE, center=TRUE)\n<\/code><\/pre>\n
\"plot<\/p>\n
clusterICE(iceGBM, nClusters = 4, plot_legend = TRUE, center=TRUE)\n<\/code><\/pre>\n
\"plot<\/p>\n
clusterICE(iceNN10, nClusters = 4, plot_legend = TRUE, center=TRUE)\n<\/code><\/pre>\n
\"plot<\/p>\n
#clusterICE(iceNN40, nClusters = 4, plot_legend = TRUE, center=TRUE)\n#clusterICE(iceNN160, nClusters = 4, plot_legend = TRUE, center=TRUE)\n<\/code><\/pre>\n
dice<\/code> : partial d<\/strong>erivative function for ice<\/code>.<\/h2>\n
dice<\/code>\ub294 ICE<\/strong> \uace1\uc120\uc744 \ubbf8\ubd84\ud55c \uacb0\uacfc\uc774\ub2e4. <\/p>\n
diceLM <- ICEbox::dice(iceLM)\nplot(diceLM)\n#plot(diceRF <- ICEbox::dice(iceRF))\n#plot(diceGAM <- ICEbox::dice(iceGAM))\n#plot(diceGBM <- ICEbox::dice(iceGBM))\nplot(diceNN10 <- ICEbox::dice(iceNN10))\n#plot(diceNN40 <- ICEbox::dice(iceNN40))\n<\/code><\/pre>\n
\"\"\/<\/p>\n
\"\"\/<\/p>\n
dice<\/code> \uace1\uc120\ub3c4 clusterICE<\/code>\ub85c \uad70\uc9d1\ud654\ud558\uc5ec \uc2dc\uac01\ud654\ud560 \uc218 \uc788\ub2e4. <\/p>\n
clusterICE(diceNN10, nClusters = 4, plot_legend = TRUE, center=TRUE)\n<\/code><\/pre>\n
\"\"\/<\/p>\n
plot<\/strong> mo<\/strong>del's residuals, response and partial dependence plots(plotmo<\/code>)<\/h2>\n
\ud328\ud0a4\uc9c0 plotmo<\/code>\uc758 plotmo<\/strong>\ub294 plot<\/strong> mo<\/strong>del's residuals, response and partial dependence plots\uc758 \uc57d\uc790\ub85c “\uadf8\ub9ac\uc790, \ubaa8\ud615\uc758 \uc794\ucc28, \ubc18\uc751(\uacb0\uacfc), \uadf8\ub9ac\uace0 \ubd80\ubd84 \uc758\uc874 \uadf8\ub9bc\uc744!"\uc774\ub77c\uace0 \uc21c\uc9c1\uc5ed[1]\ud560<\/sup> \uc218 \uc788\uaca0\ub2e4.<\/p>\n
[1]:<\/sup> \uc21c\uc11c \uadf8\ub300\ub85c\uc758 \uc9c1\uc5ed\uc774\ub77c\ub294 \uc758\ubbf8\ub85c \uc37c\ub2e4. \uc0ac\uc804\uc5d4 \uc5c6\ub294 \ub2e8\uc5b4\uc774\ub2e4.<\/p>\n
library(plotmo) # install.packages('plotmo')\n\nfitLM <- lm(y ~ ., data=dat) ## linear model\nplotmo(fitLM)\n<\/code><\/pre>\n
##  plotmo grid:    x1           x2 x3         x4         x5         x6\n##           0.1020341 -0.004727961  1 0.03112162 0.02001667 0.01247897\n##          x7         x8         x9         x10\n##  0.02389123 0.05383556 -0.0144398 -0.08341206\n<\/pre>\n
\"plot<\/p>\n
plotmo(fitRF)\n<\/code><\/pre>\n
##  plotmo grid:    x1           x2 x3         x4         x5         x6\n##           0.1020341 -0.004727961  1 0.03112162 0.02001667 0.01247897\n##          x7         x8         x9         x10\n##  0.02389123 0.05383556 -0.0144398 -0.08341206\n<\/pre>\n
\"plot<\/p>\n
#plotmo(fitRF, all2=TRUE)\n# all2=TRUE : all interaction plots\nplotmo(fitRF, all2=2)\n<\/code><\/pre>\n
##  plotmo grid:    x1           x2 x3         x4         x5         x6\n##           0.1020341 -0.004727961  1 0.03112162 0.02001667 0.01247897\n##          x7         x8         x9         x10\n##  0.02389123 0.05383556 -0.0144398 -0.08341206\n<\/pre>\n
\"plot<\/p>\n
#fitGBM <- gbm(y ~ ., data=dat, dist="gaussian", inter=2, n.trees=1000)\nplotmo(fitGBM)\n<\/code><\/pre>\n
##  plotmo grid:    x1           x2 x3         x4         x5         x6\n##           0.1020341 -0.004727961  1 0.03112162 0.02001667 0.01247897\n##          x7         x8         x9         x10\n##  0.02389123 0.05383556 -0.0144398 -0.08341206\n<\/pre>\n
\"plot<\/p>\n
#plotmo(fitGBM, i.var=1) \n#plotmo(fitGBM, i.var=c(1,2))\n\n#fitGAM <- gam(y ~ s(x1)+s(x2)+x3+s(x4)+s(x5)+s(x6)+s(x7)+s(x8)+s(x9)+s(x10),\n#              data=dat) # x3 is a categorical variable\nplotmo(fitGAM) \n<\/code><\/pre>\n
##  plotmo grid:    x1           x2 x3         x4         x5         x6\n##           0.1020341 -0.004727961  1 0.03112162 0.02001667 0.01247897\n##          x7         x8         x9         x10\n##  0.02389123 0.05383556 -0.0144398 -0.08341206\n<\/pre>\n
\"plot<\/p>\n
library(nnet) ## nnet\nset.seed(4)\n#fitNN <- nnet(y ~ ., size=20, data=scale(dat), decay=0.01, trace=FALSE)\nplotmo(fitNN10, type="raw", all2=2) # type="raw" gets passed to predict\n<\/code><\/pre>\n
##  plotmo grid:    x1           x2 x3         x4         x5         x6\n##           0.1020341 -0.004727961  1 0.03112162 0.02001667 0.01247897\n##          x7         x8         x9         x10\n##  0.02389123 0.05383556 -0.0144398 -0.08341206\n<\/pre>\n
\"plot<\/p>\n
plot<\/strong> res<\/strong>iduals(plotres<\/code>)<\/h2>\n
plotres<\/code> \ud568\uc218\ub294 \uc880 \ub354 \uc794\ucc28\uc5d0 \uc9d1\uc911\ud55c\ub2e4. <\/p>\n
plotres(fitLM)\n<\/code><\/pre>\n
\"plot<\/p>\n
plotres(fitRF)\n<\/code><\/pre>\n
\"plot<\/p>\n
plotres(fitGBM)\n<\/code><\/pre>\n
\"plot<\/p>\n
#plotres(fitXGB)\n##Error: xgboost models do not conform to standard S3 model guidelines and \n##are thus not supported by plotmo and plotres\nplotres(fitGAM)\n<\/code><\/pre>\n
\"plot<\/p>\n
plotres(fitNN10)\n<\/code><\/pre>\n
\"plot<\/p>\n
plotres(fitNN40)\n<\/code><\/pre>\n
\"plot<\/p>\n
#plotres(fitNN160)\n<\/code><\/pre>\n
\uc120\ud615 \ud68c\uadc0 \uacb0\uacfc\uc5d0 \ub300\ud574 plot<\/code>\uc744 \ud558\uace0 \uc794\ucc28\uac00 \uac00\uc815\uacfc \ubd80\ud569\ud558\ub294\uc9c0, \ubaa8\ud615\uc774 \uc5bc\ub9c8\ub098 \uc801\ud569\ud55c\uc9c0 \uace0\ubbfc\uc744 \ud574\ubcf8 \uc0ac\ub78c\uc774\ub77c\uba74 \uc704\uc758 \uadf8\ub9bc\uc5d0\uc11c\ub3c4 \ub3d9\uc77c\ud55c \ub0b4\uc6a9\uc744 \ud655\uc778\ud560 \uc218 \uc788\ub2e4. \ud2b9\ud788 \ub9c8\uc9c0\ub9c9 \uadf8\ub9bc\uc778 Residual QQ\ub97c \ubcf4\uc790. \ud788\ub4e0 \ub178\ub4dc 10\uac1c\uc9dc\ub9ac \uc778\uacf5\uc2e0\uacbd\ub9dd\uc740 \uc774\uc0c1\uc5d0 \uac00\uae5d\ub2e4! \uadf8\uc5d0 \ubc18\ud574 \ub178\ub4dc 40\uac1c\uae4c\uc9c0\ub294 \uacfc\uc801\ud569\uc774 \uc774\ub8e8\uc5b4\uc9c4 \uac83 \uac19\ub2e4. gbm<\/code> \uacb0\uacfc \uc5ed\uc2dc \uacfc\uc801\ud569\uc774 \uc774\ub8e8\uc5b4\uc9c0\uace0 \uc788\uc74c\uc744 \uccab\ubc88\uc9f8 \uadf8\ub9bc\uc5d0\uc11c \ud655\uc778\ud560 \uc218 \uc788\ub2e4. <\/p>\n
\uc2e4\uc81c\ub85c \uc0c8\ub85c\uc6b4 \ub370\uc774\ud130 \uc608\uce21\uc744 \ud1b5\ud574 \uacfc\uc801\ud569 \uc5ec\ubd80\ub97c \ud310\uc815\ud574 \ubcf4\uc790. \uba3c\uc800 NN10<\/code>\uacfc NN40<\/code>\uc758 MSE\uc744 \ube44\uad50\ud574\ubcf4\uc790. NN40<\/code>\uc740 \uae30\uc874\uc758 \ub370\uc774\ud130\uc5d0\uc11c MSE\uac00 \uc791\uc9c0\ub9cc, \uc0c8\ub85c\uc6b4 \ub370\uc774\ud130\uc5d0 \ub300\ud574\uc11c\ub294 MSE\uac00 \ucee4\uc11c \uacfc\uc801\ud569\uc774 \uc774\ub8e8\uc5b4\uc9d0\uc744 \ud655\uc778\ud560 \uc218 \uc788\ub2e4. <\/p>\n
MSEa <- mean((predict(fitNN10) - dat$y)^2)\nnewdat <- genData(iseed=101)\nMSEb <- mean((predict(fitNN10, newdata=newdat) - newdat$y)^2)\n\nMSEc <- mean((predict(fitNN40) - dat$y)^2)\nnewdat <- genData(iseed=101)\nMSEd <- mean((predict(fitNN40, newdata=newdat) - newdat$y)^2)\n\nMSE <- matrix(c(MSEa, MSEb, MSEc, MSEd), ncol=2)\ncolnames(MSE) = c('NN10', 'NN40')\nrownames(MSE) = c('\uae30\uc874 \ub370\uc774\ud130', '\uc0c8\ub85c\uc6b4 \ub370\uc774\ud130')\nMSE\n<\/code><\/pre>\n
##                    NN10      NN40\n## \uae30\uc874 \ub370\uc774\ud130   0.8994276 0.1766384\n## \uc0c8\ub85c\uc6b4 \ub370\uc774\ud130 1.3798707 3.3807835\n<\/pre>\n
\ub2e4\uc74c\uc758 \uadf8\ub798\ud504\uc5d0\uc11c \uacfc\uc801\ud569\uc758 \uc774\uc720\ub97c \uc880 \ub354 \uc0b4\ud3b4\ubcfc \uc218 \uc788\ub2e4. NN40<\/code>\uc758 \uacbd\uc6b0 \uc2e4\uc81c \\(y\\) \uac12\uc774 \ud074 \ub54c \uc608\uce21\uc774 \ud2b9\ud788 \ubd80\uc815\ud655\ud568\uc744 \ubcfc \uc218 \uc788\ub2e4. <\/p>\n
df1 = data.frame(yhat = predict(fitNN10, newdata=newdat), \n                 y=newdat$y, \n                 model='NN10')\ndf2 = data.frame(yhat = predict(fitNN40, newdata=newdat), \n                 y=newdat$y, \n                 model='NN40')\n\nlibrary(patchwork) #devtools::install_github("thomasp85\/patchwork")\n\ndf <- rbind(df1, df2)\n\nggp1 <- \n  df %>% ggplot(aes(x=y, y=yhat, col=model, shape=model)) + \n    geom_point(alpha=0.2) +\n    geom_abline(slope=1, intercept=0, linetype='dotted')\n\nggp2 <- \n  df %>% ggplot(aes(x=y, y=yhat, col=model, shape=model)) + \n    geom_point(alpha=0.2) +\n    geom_abline(slope=1, intercept=0, linetype='dotted') +\n    facet_grid(~ model)\n\nggpBlank <-   df %>% ggplot(aes(x=y, y=yhat, col=model)) + \n    geom_blank() + theme_void()\n\n(ggp1 + ggpBlank)\/ggp2 +\n  plot_annotation(title='Prediction for new data : NN10 & NN40')\n<\/code><\/pre>\n
\"plot<\/p>\n
#(ggp1 + ggpBlank + plot_layout(ncol=2)) + ggp2 + plot_layout(ncol=1)\n<\/code><\/pre>\n
\ub9c8\ubb34\ub9ac<\/h2>\n
\ubaa8\ud615\uc744 \uc124\uba85\ud558\ub294 \ubc29\ubc95, \ud328\ud0a4\uc9c0, \ud568\uc218\ub4e4\uc774 \uc815\ub9d0 \ub9ce\ub2e4. \uc544\uc9c1\ub3c4 \ub0a8\uc544 \uc788\ub2e4!<\/p>\n
\uc5ec\uae30\uc11c \uc18c\uac1c\ud55c \ud328\ud0a4\uc9c0\uc640 \ud568\uc218\ub294 \ub2e4\uc74c\uacfc \uac19\ub2e4.<\/p>\n
    \n
  • ICEbox<\/code> :: ice<\/code>, dice<\/code>, clusterICE<\/code> <\/li>\n
  • plotmo<\/code> :: plotmo<\/code>, plotres<\/code><\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"
    library(dplyr) library(ggplot2) library(randomForest) TURE <- TRUE \ubaa8\ud615, \uc794\ucc28 \uc5ec\uae30\uc11c\ub294 ICEbox \ud328\ud0a4\uc9c0\uc640 plotmo \ud328\ud0a4\uc9c0\ub97c \uc0ac\uc6a9\ud558\uc5ec \ubaa8\ud615\uc744 \uc880 \ub354 \uc0b4\ud3b4\ubcf8\ub2e4. \uc608\uc81c \ubaa8\ud615 \uc5ec\uae30\uc11c\ub294 \uc9c0\ub09c \ubc88, \uadf8\ub9ac\uace0 \uc9c0\uc9c0\ub09c\ubc88\uc758 \uc608\uc81c \ubaa8\ud615\uc744 \uadf8\ub300\ub85c \uc0ac\uc6a9\ud55c\ub2e4. \ub370\uc774\ud130 \uc0dd\uc131 \ubaa8\ud615\uc740 \ub2e4\uc74c\uacfc \uac19\ub2e4. \\[y = x_1^2 + \\frac{x_2}{|x_1|+0.1} + \\Big(I(x_2<\/p>\n","protected":false},"author":1,"featured_media":1963,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[146,319],"tags":[377,375,379,376,378],"jetpack_featured_media_url":"http:\/\/ds.sumeun.org\/wp-content\/uploads\/2019\/09\/unnamed-chunk-9-5.png","_links":{"self":[{"href":"http:\/\/ds.sumeun.org\/index.php?rest_route=\/wp\/v2\/posts\/1961"}],"collection":[{"href":"http:\/\/ds.sumeun.org\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/ds.sumeun.org\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/ds.sumeun.org\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/ds.sumeun.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=1961"}],"version-history":[{"count":3,"href":"http:\/\/ds.sumeun.org\/index.php?rest_route=\/wp\/v2\/posts\/1961\/revisions"}],"predecessor-version":[{"id":1965,"href":"http:\/\/ds.sumeun.org\/index.php?rest_route=\/wp\/v2\/posts\/1961\/revisions\/1965"}],"wp:featuredmedia":[{"embeddable":true,"href":"http:\/\/ds.sumeun.org\/index.php?rest_route=\/wp\/v2\/media\/1963"}],"wp:attachment":[{"href":"http:\/\/ds.sumeun.org\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=1961"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/ds.sumeun.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=1961"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/ds.sumeun.org\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=1961"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}