npCriterion: Feature selection based on Neyman-Pearson Criterion (NPC)

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

Description

Feature selection based on Neyman-Pearson Criterion (NPC)

Usage

npCriterion(x, y, method = c("logistic", "penlog", "svm", "randomforest",
  "lda", "slda", "nb", "nnb", "ada", "tree"), enumeration = NULL,
  max_feature_size = NULL, alpha, delta = 0.05, B = 5, l0 = 0.5,
  l1 = 0.5, seed = NULL, ncores = detectCores() - 1, ...)

Arguments

x	a design matrix
y	a vector containing binary labels 0 and 1
method	base classification method logistic: Logistic regression. glm function with family = 'binomial' penlog: Penalized logistic regression with LASSO penalty. glmnet in glmnet package svm: Support Vector Machines. svm in e1071 package randomforest: Random Forest. randomForest in randomForest package lda: Linear Discriminant Analysis. lda in MASS package slda: Sparse Linear Discriminant Analysis with LASSO penalty. nb: Naive Bayes. naiveBayes in e1071 package nnb: Nonparametric Naive Bayes. naive_bayes in naivebayes package ada: Ada-Boost. ada in ada package
enumeration	a feature set generation method, which can either be 'forward', 'backward' or 'exhaustive'. Default: 'forward'
max_feature_size	an optional integer when enumeration is 'exhaustive'. When not supplied, set to be the total number of features
alpha	a numeric scalar between 0 and 1 indicating the population type I error control
delta	a numeric scalar between 0 and 1 indicating the violation rate. Default: 0.05
B	a positive integer indicating the number of random splits. Default: 5
l0	a numeric scalar between 0 and 1 indicating the proportion of leave-out class 0 data points. Default: 0.5
l1	a numeric scalar between 0 and 1 indicating the proportion of leave-out class 1 data points. Default: 0.5
seed	random seed
ncores	a positive integer that specifies the number of cores for computing. Default: number of cores - 1.
...	additional argument for base classification methods.

Value

npCriterion returns a list with the following components:

method	the base classification method
alpha	user-specified alpha value
delta	user-specified delta value
B	total number of random splits
l0	the proportion of leave-out class 0 data points
l1	the proportion of leave-out class 1 data points
featuresets_examined	when 'enumeration' = 'forward', a list of size 2 whose first component is enumeration, and the second component is a vector of the features that are sequentially included; when 'enumeration' = 'backward', a list of size 2 whose first component is enumeration, and the second component is a vector of the features that are sequentially excluded; when 'enumeration' = 'exhaustive', a list of size 2 whose first component is enumeration, and the second component is a list of matrices whose column number ranges from 1 to max_feature_size. Rows of such a matrix represent a feature set.
npc	when 'enumeration' = 'forward' or 'backward', a vector of NPC values of feature sets in featuresets_examined; when 'enumeration' = 'exhaustive', a list of vectors of NPC values computed on the feature sets in featuresets_examined
npc.sd	when 'enumeration' = 'forward' or 'backward', a vector of standard deviations of empirical type II errors of feature sets in featuresets_examined; when 'enumeration' = 'exhaustive', a list of standard deviations of empirical type II errors computed on the feature sets in featuresets_examined
npc.se	when 'enumeration' = 'forward' or 'backward', a vector of standard errors of empirical type II errors of feature sets in featuresets_examined; when 'enumeration' = 'exhaustive', a list of standard errors of empirical type II errors computed on the feature sets in featuresets_examined
err	when 'enumeration' = 'forward' or 'backward', a vector of CV errors of feature sets in featuresets_examined; when 'enumeration' = 'exhaustive', a list of vectors of CV errors computed on the feature sets in featuresets_examined
err.se	when 'enumeration' = 'forward' or 'backward', a vector of standard deviations of test errors of feature sets in featuresets_examined; when 'enumeration' = 'exhaustive', a list of standard deviations of test errors computed on the feature sets in featuresets_examined
err.se	when 'enumeration' = 'forward' or 'backward', a vector of standard errors of test errors of feature sets in featuresets_examined; when 'enumeration' = 'exhaustive', a list of standard errors of test errors computed on the feature sets in featuresets_examined
features_minNPC	a feature set with the minimal NPC value and its corresponding NPC statistics and test errors.

Author(s)

Yiling Chen, [e-mail]

References

FILL HERE

Examples

### Example1 #####
x = matrix(rnorm(20000), ncol =5)
y = rbinom(x%*%1:5,size = 1, p =0.5)
table(y)
temp1 = npCriterion(x,y,method = "logistic",
enumeration = 'forward',
max_feature_size = NULL,
alpha =  0.05,
delta = 0.05,
B = 5,
l0 = 0.5,
l1 = 0.5)

temp2 = npCriterion(x,y,method ="svm",
kernel = 'radial',
enumeration = 'exhaustive',
alpha =  0.05,
delta = 0.05,
B = 5,
l0 = 0.5,
l1 = 0.5)
### Example2 #####
y = rbinom(100000,size = 1, p =0.5)
x = matrix(NA, nrow =100000,ncol =2)
x1 = cbind(rnorm(sum(y==1),mean =1, sd =1),rnorm(sum(y==1),mean =1, sd =1))
x0 = cbind(rnorm(sum(y==0),mean =-1, sd =1),rnorm(sum(y==0),mean =0.5, sd =1.5))
pnorm(qnorm(0.95,-1,1),1,1)
pnorm(qnorm(0.95,0.5,1.5),1,1)
x[y==1,] = x1
x[y==0,] = x0
table(y)

temp3 = npCriterion(x,y,method ="lda",
                    enumeration = 'exhaustive',
                    alpha =  0.05,
                    delta = 0.05,
                    B = 5,
                    l0 = 0.5,
                    l1 = 0.5)
temp3$criteria$`ell=1`

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