uml: Mixed stock analysis by unconditional maximum likelihood

In mixstock: functions for mixed stock analysis

Description

Find the unconditional maximum likelihood estimate (jointly estimating marker frequencies in sources) of the contributions of different sources to a mixed stock, by either a direct-search or an expectation-maximization method

Usage

uml(x, method="direct",optmethod="L-BFGS-B",...)
uml.ds(x,grad=uml.grad,start.type="lsolve",fuzz=0,bounds=1e-4,
ndepfac=1000,method="L-BFGS-B",debug=FALSE,control=NULL,
transf=c("part","full","none"),...)
uml.em(x,prec=1e-8,prior=1)

Arguments

x	a list with elements mixsamp (a vector of the sampled markers in the mixed stock) and sourcesamp (a matrix, with markers in rows and sources in columns, of markers in the source samples)
optmethod	to be passed to optim
grad	function giving the gradient of the likelhood
start.type	starting values to use: equal (equal contributions from each source); random (multinomial sample with equal probabilities); rand2 (sample out of a transformed normal distribution); a number between 1 and the number of sources; that source starts with 0.95 contribution and the rest start with 0.05/(R-1); default lsolve, the linear solution to the problem
fuzz	min. value (1-min is the max.) for starting contributions
bounds	(bounds,1-bounds) are the lower and upper bounds for mle calculations
ndepfac	factor for computing numerical derivatives; numerical derivative stepsize is computed as bounds/ndepfac [OBSOLETE with gradient function?]
method	optimization method, to be passed to optim
transf	transformation
debug	produce debugging output?
control	other control arguments to optim
...	other arguments to mle or optim (e.g. hessian=FALSE to suppress (slow) hessian calculation, etc.)
prec	precision for determining convergence of EM algorithm
prior	prior for EM algorithm

Details

uml uses either a direct-search algorithm or an EM algorithm to find the ML estimate

Value

an object of class mixstock.est, with elements

fit	information on the ML fit
resample	bootstrap information, if any
data	original data used for estimate
R	number of sources
H	number of markers
contin	estimation done on transformed proportions?
method	optimization method
boot.method	resampling method
boot.data	raw resampling information
gandr.diag	Gelman-Rubin diagnostic information for MCMC estimates
prior	Prior for MCMC estimates
em	estimation done by EM algorithm?

Author(s)

Ben Bolker

Examples

true.freq <- matrix(c(0.65,0.33,0.01,0.01,
                      0.33,0.65,0.01,0.01),ncol=2)
true.contrib <- c(0.9,0.1)
x <- simmixstock0(true.freq,true.contrib,50,100,1004)
uml.est <- uml(x)
uml.est
uml.emest <- uml.em(x)
uml.emest

mixstock documentation built on May 2, 2019, 6:48 p.m.