bhd: BHD take home

#' Predict LDL levels
#'
#' This function uses LASSO to select predictors in order to run linear regression to predict LDL for individuals. 
#' Outputs a vector of predicted values of LDL corresponding on input
#' There are no checks for NAs or anythingelse and thus this isa development level function.
#' @param data Data.frame or matrix where the first column is what you want to predict, and 
#' remaining columns are what you want to consider as possibly covariates. Any factors MUST be set as factors before 
#' running. No character vectors allowed.
#' @keywords preds
#' @return A vector of length equal to number of rows of input data with predicted values of LDL
#' @export
#' @examples
#' response <- rnorm(1000)
#' goodpredictors <- 0.5*response + rnorm(1000,0.5)
#' badpredictors <- matrix(rnorm(10000),ncol=10)
#' estimate.ldl(cbind(response,goodpredictors,badpredictors))


estimate.ldl <- function(data){
  nc <- ncol(data)
  x <- as.matrix(data[,2:nc])
  m1 <- glmnet::cv.glmnet(x,data[,1])
  keep <- ((as.numeric((coef(m1, s = "lambda.1se"))))!=0)[-1]
  if(sum(keep)==0){
    return("No variables appeared to be important, no predictions made")
  }
  #xreduced <- x[,which(coef(m1, s = "lambda.1se") !=0)[-1]-1]
  xreduced <- x[,keep]
  m2 <- lm(data[,1]~xreduced)
  preds <- predict(m2)
  return(preds)
}