An example of how each column in a data frame could be residualized for some covariates

bias_correct.R

# Download example data from: https://github.com/johncolby/SVM-RFE/zipball/master 

setwd('/path/to/SVM-RFE/') # Change this to your setup

load('demo/input.Rdata')
data  = input[, 2:3]
input = input[, -(1:3)]

# Function to residualize a vector x for the covariates in data
residualize <- function(x, fun=x~., data) {
  data = cbind(x=x, data)
  lm(fun, data=data)$resid
}

# Apply our residualize function to each column of input
input = colwise(residualize)(input, data=data)

################################################################################
# Modification for test data

load('demo/input.Rdata')

# Setup training data
input.train = input[1:100, ]
data.train  = input.train[, 2:3]
input.train = input.train[, -(1:3)]

# Setup test data
input.test = input[101:nrow(input), ]
data.test  = input.test[, 2:3]
input.test = input.test[, -(1:3)]

# Modified residualize function to residualize test data based on model fit to 
# training data only
residualizeTest <- function(x, fun=x~., data, x.new, data.new) {
  data = cbind(x=x, data)
  fit = lm(fun, data=data)
  
  data.new = cbind(x=x.new, data.new)
  pred = predict(fit, data.new)
  x.new - pred
}

# Apply the function to each column in our test data
# (more complicated now so can't simply use colwise)
input.test = sapply(1:ncol(input.test), function(x) residualizeTest(input.train[,x], data=data.train, x.new=input.test[,x], data.new=data.test))