splitt {IntegratedMRF}

Split of the Parent node

Description

Split of the training samples of the parent node into the child nodes based on the feature and threshold that produces the minimum cost

Usage

splitt(X, Y, m_feature, Index, Inv_Cov_Y, Command, ff)

Arguments

X

Input Training matrix of size M x N, M is the number of training samples and N is the number of features

Y

Output Training response of size M x T, M is the number of samples and T is the number of output responses

m_feature

Number of randomly selected features considered for a split in each regression tree node.

Index

Index of training samples

Inv_Cov_Y

Inverse of Covariance matrix of Output Response matrix for MRF (Input [0 0; 0 0] for RF)

Command

1 for univariate Regression Tree (corresponding to RF) and 2 for Multivariate Regression Tree (corresponding to MRF)

ff

Vector of m_feature from all features of X. This varies with each split

Details

At each node of a regression a tree, a fixed number of features (m_feature) are selected randomly to be considered for generating the split. Node cost for all selected features along with possible n-1 thresholds for n samples are considered to select the feature and threshold with minimum cost.

Value

List with the following components:

index_left

Index of the samples that are in the left node after splitting

index_right

Index of the samples that are in the right node after splitting

which_feature

The number of the feature that produces the minimum splitting cost

threshold_feature

The threshold value for the node split. A feature value less than or equal to the threshold will go to the left node and it will go to the right node otherwise.

Examples

library(IntegratedMRF)
X=matrix(runif(20*100),20,100)
Y=matrix(runif(20*3),20,3)
m_feature=5
Index=1:20
Inv_Cov_Y=solve(cov(Y))
ff2 = ncol(X) # number of features
ff =sort(sample(ff2, m_feature)) 
Command=2#MRF, as number of output feature is greater than 1
Split_criteria=splitt(X,Y,m_feature,Index,Inv_Cov_Y,Command,ff)