README.md

In KOBT: Knockoff Boosted Tree

KOBT

The goal of KOBT is to conduct variable selection in tree models with false discovery rate control. The difference of SHAPs between original and knockoff variables is used as the test statistic.

Example

This is a basic example which shows you how to (1) generate knockoffs, and (2) conduct variable selection in tree models with false discovery rate control using KOBT.

## basic example code
library('KOBT')
# 1. Generate Knockoffs
set.seed(10)
X <- matrix(rnorm(100), nrow = 10)
Z <- generate.knockoff(X = X, type = "shrink", num = 2)
# 2. Conduct variable selection
beta <- rep(0, 10)
beta[1:5] <- 10
Y <- MASS::mvrnorm(n = 1, mu = X%*%beta, Sigma = diag(10))
result <- vector(mode = "list", length = length(Z))
for (i in 1:length(Z)) {
  x <- cbind(X, Z[[i]])
  dtrain <- xgboost::xgb.DMatrix(x, label = Y)
  fit.model <- xgboost::xgb.train(data = dtrain, nrounds = 2)
  result[[i]] <- importance.score(fit = fit.model, Y = Y, X = x)$shap
}
output <- matrix(unlist(result), ncol = length(result[[1]]), byrow = TRUE)
selected.index <- kobt.select(score = output)

KOBT documentation built on Feb. 20, 2020, 5:08 p.m.