R/extract_coef.R

In sqlscore: Utilities for Generating SQL Queries from Model Objects

# Standardized coefficient extraction
#
# Extract model coefficients in a standardized format (as a named numeric vector).
#
# @section Warning:
# The Binomial models in glmboost return coefficients which are 1/2 the coefficients
# fit by a call to glm(..., family=binomial(...)), because the response variable is
# internally recoded to -1 and +1. sqlscore multiplies the returned coefficients by 2
# to put them back on the same scale as glm, and adds the glmboost offset to the
# intercept before multiplying.
#
# @param object An object for which the extraction of model coefficients is meaningful.
#
# @return Model coefficients as a named numeric vector.
extract_coef <-
function(object)
{
  UseMethod("extract_coef")
}

#' @export
extract_coef.default <-
function(object)
{
  stats::coef(object)
}

#' @export
extract_coef.glmboost <-
function(object)
{
  # suppress coef.glmboost's message so we can print our own
  cf <- captureConditions(stats::coef(object, off2int=TRUE))$value

  # mboost internally recodes DVs to -1 and +1 in one particular case (the
  # logit fit from Binomial_adaboost()), so the coefficients are half those
  # returned by glm. If we have a binomial model, let's fix this and return
  # twice the fitted coefficients, with a message
  if(object$family@name == "Negative Binomial Likelihood (logit link)")
  {
    message("\nNOTE: Coefficients from glmboost's Binomial_adaboost logit model \n",
            "are 1/2 the coefficients from a model fit by glm(... , family = 'binomial').\n",
            "sqlscore scales these coefficients by 2 to put them on the same scale as glm.\n")
    sc <- 2
  }
  else
  {
    sc <- 1
  }

  sc * cf
}

#' @export
extract_coef.cv.glmnet <-
function(object)
{
  cf <- stats::coef(object)

  val <- as.vector(cf)
  names(val) <- rownames(cf)

  #Return only the coefficients that weren't regularized to 0
  val[which(val != 0)]
}