gbm_foldid_eg.R

library("devtools")
install_github("DexGroves/gbm")
library("gbm")

# Data generation from gbm manual ----------------------------------------------
set.seed(112358)
N     <- 1000
X1    <- runif(N)
X2    <- 2*runif(N)
X3    <- ordered(sample(letters[1:4],N,replace=TRUE),levels=letters[4:1])
X4    <- factor(sample(letters[1:6],N,replace=TRUE))
X5    <- factor(sample(letters[1:3],N,replace=TRUE))
X6    <- 3*runif(N)
mu    <- c(-1,0,1,2)[as.numeric(X3)]
SNR   <- 10
Y     <- X1**1.5 + 2 * (X2**.5) + mu
sigma <- sqrt(var(Y)/SNR)
Y     <- Y + rnorm(N,0,sigma)
X1[sample(1:N,size=500)] <- NA
X4[sample(1:N,size=300)] <- NA
data <- data.frame(Y=Y,X1=X1,X2=X2,X3=X3,X4=X4,X5=X5,X6=X6)

# Fit some models without fold.id ----------------------------------------------
set.seed(123456)
g1 <- gbm(Y ~ X1 + X2 + X3 + X4 + X5 + X6,
            data              = data,
            distribution      = "gaussian",
            n.trees           = 2000,
            bag.fraction      = 1.0,
            shrinkage         = 0.01,
            interaction.depth = 3,
            cv.folds          = 3,
            n.cores           = 4)

set.seed(260591)
g2 <- gbm(Y ~ X1 + X2 + X3 + X4 + X5 + X6,
            data              = data,
            distribution      = "gaussian",
            n.trees           = 2000,
            bag.fraction      = 1.0,
            shrinkage         = 0.01,
            interaction.depth = 3,
            cv.folds          = 3,
            n.cores           = 4)

# Fit some models with fold.id -------------------------------------------------
set.seed(123456)
folds <- sample(seq(1,3), nrow(data), replace = TRUE)

set.seed(123456)
gid1 <- gbm(Y ~ X1 + X2 + X3 + X4 + X5 + X6,
            data              = data,
            distribution      = "gaussian",
            n.trees           = 2000,
            bag.fraction      = 1.0,
            shrinkage         = 0.01,
            interaction.depth = 3,
            fold.id           = folds,
            n.cores           = 4)

set.seed(260591)
gid2 <- gbm(Y ~ X1 + X2 + X3 + X4 + X5 + X6,
            data              = data,
            distribution      = "gaussian",
            n.trees           = 2000,
            bag.fraction      = 1.0,
            shrinkage         = 0.01,
            interaction.depth = 3,
            fold.id           = folds,
            n.cores           = 4)

# g1 and g2 are seed-dependent, gid1 and gid2 are not
gbm.perf(g1)
gbm.perf(g2)
gbm.perf(gid1)
gbm.perf(gid2)