Convert data frame with factors into one with dummy variables for all the factors. Also take care of missing values etc.

gistfile1.r

## build a nice design matrix
## formula can be a vector of column names or a standard formula
design <- function(formula, response, data){
  if(is.character(formula)){
    data <- data[,formula]
    formula <- ~.
  }
  if(is.null(dim(data))) data <- as.matrix(data)
  if(is.null(rownames(data))) rownames(data) <- 1:nrow(data)
  response <- c(response)
  if(nrow(data)!=length(response))
    stop("response must have same number of observations as the data")
  names(response) <- rownames(data)
  data <- as.data.frame(data)
  x <- as.data.frame(model.matrix(formula, data=data))[,-1]
  y <- response[rownames(x)]
  x <- x[!is.na(y),]
  y <- y[!is.na(y)]
  return(list(X=x,Y=y,formula=formula)) 
}

## mean imputation for a numeric design matrix
meanimpute <- function(x){
  x$MISSING <- apply(x,1,function(r) as.numeric(any(is.na(r))))
  x <- as.data.frame(apply(x,2,
                           function(c){
                             avg <- mean(c,na.rm=TRUE)
                             c[is.na(c)] <- avg
                             return(c) }))
  return(x) }