fit.4g: fit.4g

In jamesliley/cfdr: Implements cFDR testing with overall FDR control

Description

Fit a four-part mixture normal model to bivariate data. Assumes that data are distributed as one of N(0,1) x N(0,1) with probability pi0 N(0,s1^2) x N(0,1) with probability pi1 N(0,1) x N(0,t1^2) with probability pi2 N(0,s2^2) x N(0,t2^2) with probability 1-pi0-pi1-pi2 Fits the set of parameters (pi0,pi1,pi2,s1,s2,t1,t2) using an E-M algorithm

Usage

fit.4g(P, pars = c(0.7, 0.1, 0.1, 2, 2, 2, 2), weights = rep(1,
  dim(Z)[1]), C = 1, maxit = 10000, tol = 1e-04, sgm = c(1, 100))

Arguments

P	matrix N x 2 of data points (Z scores or P-values)
pars	initial parameter values
weights	weights for points
C	include additive term C*log(pi0*pi1*pi2*(1-pi0-pi1-pi2)) in objective function to ensure identifiability of model
maxit	maximum number of iterations (supersedes tol)
tol	stop after increment in log-likelihood is smaller than this
sgm	force s1,s2,t1,t2 to be at least this value

Value

list with elements pars (fitted parameters), lhood (log likelihood) and hist (fitted parameters during algorithm

Author(s)

James Liley

Examples

pi0=0.5; pi1=0.15; pi2=0.25
s1=3; s2=2; t1=2; t2=3
true_pars=c(pi0,pi1,pi2,s1,s2,t1,t2)


n=100000; n0=round(pi0*n); n1=round(pi1*n); n2=round(pi2*n); n3=n-n0-n1-n2

zs=c(rnorm(n0,sd=1),rnorm(n1,sd=s1),rnorm(n2,sd=1),rnorm(n3,sd=s2))
zt=c(rnorm(n0,sd=1),rnorm(n1,sd=1),rnorm(n2,sd=t1),rnorm(n3,sd=t2))

Z=cbind(zs,zt)

f=fit.4g(Z)
f$pars

jamesliley/cfdr documentation built on July 31, 2020, 9:42 a.m.