BMhyb (version 1.5.2)

CalculateLikelihood: Calculate the likelihood value for the model

Description

This function calculates the likelihood value for the model.

Usage

CalculateLikelihood(x, data, phy, flow, actual.params, precision=2,
	proportion.mix.with.diag=0, allow.extrapolation=FALSE,
	measurement.error = NULL, do.kappa.check = FALSE, number.of.proportions=101,
	lower.b=c(0, -Inf, 0.000001, 0, 0), upper.b=c(10,Inf,100,100,100),
	badval.if.not.positive.definite=TRUE, do.Brissette.correction=FALSE, 
	do.Higham.correction=TRUE, do.DE.correction=FALSE, return.penalty=FALSE, ...)

Arguments

x

parameter of interest.

data

the trait values.

phy

a tree of phylo class.

flow

strcuture of gene flow.

actual.params

the free parameters for hybridization.

precision

a number to verify the condition of the variance covariation for the network model.

proportion.mix.with.diag

the proportion value that applies to the diagonal of the vcv matrix.

allow.extrapolation

a TRUE/FALSE argument.

measurement.error

estimate or fixed (see details).

do.kappa.check

check for VCV matrix condition, approximate if poor. Leave TRUE unless you're cleverer than us.

number.of.proportions

control how many points to use for extrapolation for ill-conditioned matrices.

lower.b

minimum values of all possible parameters (non-free ones will be deleted).

upper.b

maximum values of all possible parameters (non-free ones will be deleted).

badval.if.not.positive.definite

If the VCV is not positive definite, return a large bad value for likelihood.

do.Brissette.correction

If VCV is not positive definite, do Brissette et al. (2007) correction.

do.Higham.correction

If VCV is not positive definite, do Higham (2002).

do.DE.correction

If VCV is not positive definite, find a nearby matrix that is.

return.penalty

If TRUE, return a penalty for approximating not the likelihood itself.

...

place to absorb other arguments.

Value

negative log likelihood value.

Details

This function calculates the likelihood value. As described in the argument, the \(x\) contains the parameter of interests including over all mean \(\mu\), the rate of evolution \(\sigma^2\), the measurement error SE and the hybrid vigor \(\beta\) and the variation at the burst of hybridization \(v_H\). Names of the taxa in data vector and the phylogeny must be match for further analysis, otherwise it would terminate immediately. The negative log likelihood function is calculated and a reasonable likelihood value is returned by checking the variance covariance matrix where a precision number is used for verifying the log conditioned number of the variance covariance of the network model. When the matrix is ill conditioned, we modestly adjust the matrix by the shrink the value of the off diagonal matrix using a proportion array. Then the likelihood is calculated using a spline approximation.

References

Jhwueng D.C. and O'Meara B.C. 2016. Studying trait evolution on phylogenetic networks. Submitted.

Examples

# NOT RUN {
	#set the number of hybrid
	ntax.nonhybrid<-5
	#set the number of hybrid
	ntax.hybrid<-1
	#simulate a network
	network<-SimulateNetwork(ntax.nonhybrid=ntax.nonhybrid, ntax.hybrid=ntax.hybrid,
	flow.proportion=0.5, origin.type='clade', birth = 1, death = 0.5, sample.f = 0.5,
	tree.height = 1, allow.ghost=FALSE)
sigma.sq <- 0.01;mu <- 1;SE <- 0
	#simulatedata
	data<-rnorm(ntax.nonhybrid+ntax.hybrid)
	names(data)<-paste("t",(1:(ntax.nonhybrid+ntax.hybrid)),sep="")
	#calculate the likelihood value
	CalculateLikelihood(c(sigma.sq,mu,SE), data, network$phy, network$flow, "vh",
	precision=2, proportion.mix.with.diag=0, allow.extrapolation=TRUE)
	
# }