sauc.phi: sauc.phi

In aucm: AUC Maximization

Description

sauc.phi optimizes Normal CDF approximation of AUC using Newton Raphson

Usage

sauc.phi (formula,dat,constrain.method="L2",h.method="Lin",start.method="rlogit",
opt.method = "Lin", upper = NULL, verbose = FALSE, ret.vcov = FALSE, truth = NULL, 
    beta.init=NULL)

Arguments

formula	a formula
dat	a data frame
constrain.method	a string. If "L2", L2 norm is constrained to 1. If "beta1", beta1 is fixed to 1. Default "L2".
h.method	a string. If "Lin", Lin et al, data dependent. If "Vexler", (n1*n2)^(-0.1) Vexler et al (2006). If "MH", Ma and Huang. Default "Lin".
start.method	a string. If "rlogit", robust logistic fit is used as beta.init If "1", a vector of 1 is used as beta.init. Default "rlogit".
opt.method	character string, possible values are "truth","YH","Lin", please see code for more details
upper	required for opt.method = 'YH'
verbose	logical
ret.vcov	logical, whether to return an estimate of the covariance matrix of 'beta' for normal or logistic sigmoid functions.
truth	numeric, it will be returned as the result of the fit, please see code for more details
beta.init	vector. Initial values for coefficients.

Details

If an error happens during optimization (typically due to solve()), the errors are catched and NAs are returned.

Author(s)

Shuxin Yin
Ying Huang
Youyi Fong [e-mail]

Examples

seed=26
seed=16
seed=3
dat.train = sim.dat.1(n=200, seed=seed, add.outliers=TRUE) 
fits=list()
fits[[1]]=sauc.phi(y~x1+x2, dat.train,constrain.method="L2",h.method="Lin")
fits[[2]]=sauc.phi(y~x1+x2, dat.train,constrain.method="L2",h.method="MH")
fits[[3]]=sauc.phi(y~x1+x2, dat.train,constrain.method="beta1",h.method="Lin")
fits[[4]]=sauc.phi(y~x1+x2, dat.train,constrain.method="beta1",h.method="MH") 
# not a good combination of constrain.method and h.method
sapply(fits, function(x) ratio(x)[2])


# explosion
seed=954
dat.train = sim.dat.1(n=200, seed=seed, add.outliers=TRUE) 
fit.1 = sauc.phi(y~x1+x2, dat.train,constrain.method="L2",h.method="Lin")
ratio(fit.1)

aucm documentation built on Jan. 11, 2020, 9:43 a.m.