variable_importance_measure {MultivariateRandomForest}

Calculates variable Importance of a Regression Tree Model

Description

Number of times a variable has been picked in the branch nodes of a (single) regression tree.

Usage

variable_importance_measure(Model_VIM,NumVariable)

Arguments

Model_VIM

Regression Tree model in which the variable importance is measured

NumVariable

Number of variables in the training or testing matrix

Details

In time of calculating node cost of a tree of a random forest, a user defined number of variables are randomly picked. Among this, the best variable is chosen for the node using the node cost. While an important variable for a model will always come out as the best. This function calculates the number of times a variable has been picked in the regression tree. It has been done by checking which variables are picked, how many times, in the branch nodes of the model.

Value

Vector of size (1 x NumVariable), showing the number of repetition of variables (serially) in the branch nodes of the model.

Examples

library(MultivariateRandomForest)
trainX=matrix(runif(50*100),50,100)
trainY=matrix(runif(50*5),50,5) 
n_tree=2
m_feature=5
min_leaf=5
testX=matrix(runif(10*100),10,100) 

theta <- function(trainX){trainX}
results <- bootstrap::bootstrap(1:nrow(trainX),n_tree,theta) 
b=results$thetastar

Variable_number=ncol(trainY)
if (Variable_number>1){
  Command=2
}else if(Variable_number==1){
  Command=1
} 
NumVariable=ncol(trainX)
NumRepeatation=matrix(rep(0,n_tree*NumVariable),nrow=n_tree)

for (i in 1:n_tree){
  Single_Model=NULL
  X=trainX[ b[ ,i],  ]
  Y=matrix(trainY[ b[ ,i],  ],ncol=Variable_number)
  Inv_Cov_Y = solve(cov(Y)) # calculate the V inverse
  if (Command==1){
    Inv_Cov_Y=matrix(rep(0,4),ncol=2)
  }
  Single_Model=build_single_tree(X, Y, m_feature, min_leaf,Inv_Cov_Y,Command)
  NumRepeatation[i,]=variable_importance_measure(Single_Model,NumVariable)
}