README.md

In yiling0210/NPCriterion: Neyman-Pearson Criterion (NPC): A Model Selection Criterion for Asymmetric Binary Classification

NPCriterion

Model comparison criterion under the Neyman Pearson binary classification paradigm

Introduction

NPCriterion is proposed to compare models under the Neyman Pearson(NP) binary classification paradigm. Users can refer to our paper Neyman-Pearson Criterion (NPC): A Model Selection Criterion for Asymmetric Binary Classification for a detailed description of the criterion and applications.

Any suggestions on the package are welcome! For technical problems, please report to Issues. For suggestions and comments on the method, please contact Yiling([e-mail]) or Dr. Jessica Li ([e-mail]).

Installation

The package is not on CRAN yet. For installation please use the following codes in R

install.packages("devtools")
library(devtools)

install_github("yiling0210/NPCriterion")

Quick start

NPCriterion is built to integrate many common classification methods, including logistic regression, penalized logistic regression, svm, random forest, linear discriminant analysis (lds), sparse lda, naive Bayes, nonparametric Naive Bayes and adaBoost. Its mandatory inputs include the covariates matrix x, response y, classification method method and the type I error control alpha:

x = matrix(rnorm(20000), ncol =5)
y = rbinom(x%*%1:5,size = 1, p =0.5)
table(y)


npCriterion(x = x,  # covariate matrix 
            y = y,  # binary vector response
            method = "logistic", # classification method
            enumeration = "forward", # forward
            alpha = 0.05, # type I error control
            delta = 0.05, # violation rate
            B = 5, # number of random splits
            l0 = 0.5, # proportion of leave-out class 0 data
            l1 = 0.5, # proportion of leave-out class 1 data
            ncores = detectCores() - 1 # number of cores for parallel computing
            )

npCriterion returns a list, including features that have been examined featuresets_examined and a feature set with the minimal NPC value features_minNPC

Sometimes due to computational power limits, users may want to do feature screening first:

n = 1000; p = 1000
y = rbinom(n,size = 1, p =0.5)
x = matrix(NA, nrow =n,ncol =p)
x1 = rmvnorm(sum(y==1), mean = rep(1, p), sigma = diag(1,p))
x0 = rmvnorm(sum(y==0), mean = seq(from = -2, to = 1, length.out = p), sigma = diag(1,p))
x[y==1,] = x1
x[y==0,] = x0
table(y)

npMarginal(x = x, 
           y = y, 
           method = "logistic",
            p.adjust.methods = "BH",
            N = 50, 
            alpha = 0.05, 
            epsilon = 0.1,
            l0 = 0.5, 
            l1 = 0.5,
            )

For detailed usage, please refer to the package manual.

yiling0210/NPCriterion documentation built on May 10, 2019, 1:25 p.m.