kristinblesch / cpi

cpi

Conditional Predictive Impact

David S. Watson, Marvin N. Wright

Introduction

Conditional Predictive Impact (CPI) is a general test for conditional independence in supervised learning algorithms. The measure can be calculated using any supervised learning algorithm and loss function. It provides statistical inference procedures without parametric assumptions and applies equally well to continuous and categorical predictors and outcomes.

Installation

The package is not on CRAN yet. To install the development version from GitHub using devtools, run

devtools::install_github("dswatson/cpi")

Examples

Calculate CPI for random forest on iris data with 5-fold cross validation:

mytask <- makeClassifTask(data = iris, target = "Species")
cpi(task = mytask, 
    learner = makeLearner("classif.ranger", num.trees = 50),
    resampling = makeResampleDesc("CV", iters = 5), 
    measure = "mmce", test = "t")

References

• Watson D. S. & Wright, M. N. (2018). Testing conditional independence in supervised learning algorithms. In preparation.