importance.forestRK {forestRK}

Calculates Gini Importance or Mean Decrease Impurity (same algorithm is used in 'scikit-learn') of each covariate that we consider in the forestRK model

Description

Calculates Gini Importance of each covariate considered in the forestRK model, and list them in the order of most important to the least important.

The Gini Importance or Mean Decrease in Impurity algorithm is also used in 'scikit-learn'. Gini Importance is defined as the total decrease in node impurity averaged over all trees of the ensemble, where the decrease in node impurity is obtained after weighting by the probability for an observation to reach that node (which is approximated by the proportion of samples reaching that node).

Usage

 importance.forestRK(forestRK.object = forestRK())

Arguments

forestRK.object

a forestRK object.

Value

A list containing the following items:

importance.covariate.names

a vector of names of the covariates of our dataset ordered from the most important to the least important.

average.decrease.in.criteria.vec

a vector storing the average decrease in the weighted splitting criteria by each covariate that was calculated across all trees in the forestRK object; the numbers are ordered from the highest average decrease in criteria (importance) to the lowest, so the i-th importance number from this vector pertains to the i-th covariate listed in the vector output importance.covariate.names.

ent.status

status of the parameter entropy; TRUE if Entropy is used for splitting critera, FALSE if Gini Index is used instead.

x.original

a numericized data frame storing covariates of each observation from the given (training) dataset that was used to construct the forestRK object in the beginning of the forestRK function call.

Author(s)

Hyunjin Cho, [e-mail] Rebecca Su, [e-mail]

See Also

forestRK

Examples

  ## example: iris dataset
  ## load the forestRK package
  library(forestRK)

  ## numericize the data
  x.train <- x.organizer(iris[,1:4], encoding = "num")[c(1:25,51:75,101:125),]
  y.train <- y.organizer(iris[c(1:25,51:75,101:125),5])$y.new

  # random forest
  # min.num.obs.end.node.tree is set to 5 by default;
  # entropy is set to TRUE by default
  # typically the nbags and samp.size has to be much larger than 30 and 50
  forestRK.1 <- forestRK(x.train, y.train, nbags=30, samp.size=50)
  # execute importance.forestRK function
  imp <- importance.forestRK(forestRK.1)