R/dg_batch_seq.R

In Beuleup93/dgrGlm: Regression logistic binary tools

#' Gradient descent algorithm
#'
#' @param X is the matrix of our predictor variables with the bias column
#' @param y is a target variable to predict.
#' @param theta is a vector containing the parameters or coefficient of the logistic to be estimated
#' @param leaning_rate is the learning rate that controls the magnitude of the vector update.
#' @param max_iter is the number of iterations.
#' @param tolerance an additional parameter which specifies the minimum movement allowed for each iteration
#' @param rho hyper parameter which allows arbitration between RDIGE and LASSO.
#' @param C parameter allowing to arbitrate between the penalty and the likelihood in the guidance of the modeling.
#' @importFrom stats na.omit
#'
#'
#' @export
#' @author "Saliou NDAO <salioundao21@gmail.com>"
#' @return this function returns the instance of model with all parameters
#'
#' @examples
#' \dontrun{
#'   gradient(X,y,theta)
#'   gradient(X,y,theta, leaning_rate=0.1, max_iter=100, tolerance=1e-04)
#' }
dg_batch_seq<- function(X,y,theta, leaning_rate, max_iter, tolerance, rho=NA, C=NA){

  if (dim(X)[1] != length(y)){
    stop("the dimensions of 'x' and 'y' do not match")
  }
  # FOR INSTANCE CLASS
  instance <- list()

  # FOR COST HISTORISATION AND ITERATION
  cost_vector = c()
  iter <- 0

  while(iter < max_iter){
    iter <- iter + 1
    if(is.na(C) && is.na(rho)){
      cost = logLoss(theta, as.matrix(X), y)
      grad = gradient(theta, as.matrix(X), y)
    }else{
      cost = logLossElasticnet(theta, as.matrix(X), y, rho, C)
      grad = gradientElasticnet(theta, as.matrix(X), y, rho, C)
    }
    cost_vector = c(cost_vector, cost)
    new_theta = theta - leaning_rate*grad
    if (sum(abs(new_theta-theta)) < tolerance){
      break
    }
    theta = new_theta
  }
  instance$theta <- theta
  instance$history_cost <- cost_vector
  instance$nbIter <- iter
  class(instance) <- "gradient_desc_batch"
  return(instance)
}