permuteRelief {AppliedPredictiveModeling}

Permutation Statistics for the Relief Algorithm

Description

This function uses a permutation approach to determining the relative magnitude of Relief scores (Kira and Rendell, 1992 and Kononenko, 1994).

Usage

permuteRelief(x, y, nperm = 100, ...)

Arguments

x

a data frame of predictor data

y

a vector of outcomes

nperm

the number of random permutations of the data

...

options to pass to attrEval, such as the exact Relief algorithm, to use

Details

The scores for each predictor are computed using the original data and after outcome data are randomly scrambled (nprem times). The mean and standard deviation of the permuted values are determined and a standardized version of the observed scores are determined by subtracting the permuted means from the original values, then dividing each by the corresponding standard deviation.

Value

a list with elements

standardized

a vector of standardized predictor scores

permutations

the values of the permuted scores, for plotting to assess the permutation distribution

observed

the observed scores

options

a list of options passed using ...

Author(s)

Max Kuhn

References

Kira, K., & Rendell, L. (1992). The feature selection problem: Traditional methods and a new algorithm. Proceedings of the Eleventh International Conference on Machine Learning, 129-129.

Kononenko, I. (1994). Estimating attributes: analysis and extensions of RELIEF. Machine Learning: ECML-94, 171-182.

See Also

attrEval

Examples

set.seed(874)
reliefEx3 <- easyBoundaryFunc(500)
reliefEx3$X1 <- scale(reliefEx3$X1)
reliefEx3$X2 <- scale(reliefEx3$X2)
reliefEx3$prob <- NULL

standardized <- permuteRelief(reliefEx3[, 1:2], reliefEx3$class, 
                              ## For efficiency, a small number of
                              ## permutations are used here.
                              nperm = 50,           
                              estimator="ReliefFequalK", 
                              ReliefIterations= 50)