optmod: MLEs for a given model.

In lnar: Inference for stochastic kinetic genetic networks.

Description

Fits the compiled model to a given datased using a numerical Maximul Likelihood Estimation procedure.

Usage

optmod(cout,nthetas, mydata, maxiter=300,
       syssize=sum(mydata[1,-1]), tcrit=.0001,
       relerr=1e-9, abserr=1e-9, hessianh=1e-4,
       method=1, usebfgs=0,hess=1, dfunction)

Arguments

cout	The parsed model containing the C code and the name relations.
nthetas	A numerical vector with the initial values for the parameters to be optimized.
mydata	A data.frame or a matrix with the data. The first column must indicate the time of the observations.
maxiter	Numerical, indicated the maximum number of iterations for the optimization algorithm.
syssize	Numerical, the system size defaults to the initial population.
tcrit	Numerical, the convergence criterion for the optimization algorithm.
relerr	Numerical, the relative error for the numerical ordinary differential equations (ODEs) solver.
abserr	Numerical, the absolute error for the numerical ordinary differential equations (ODEs) solver.
hessianh	Numerical, indicates the approximation step for the central differences calculations of the Hessian matrix.
method	Numerical which takes the following integer values: 0: Restarting Method using number of molecules. 1: Restarting Method using concentrations. 3: Non-Restarting Method using concentrations. a discussion on the Restarting and the Non Restarting method can be found in Giagos (2010).
usebfgs	Specify whether to use the BFGS algorithm (1), or the default Nelder-Mead simplex algorithm (0)
hess	Specify whether to calculate the Hessian matrix (1), or not (0)
dfunction	The compiled function, given as a loaded dynamic library in R.

Details

By default the L-BFGS-B optimization procedure is employed (see optim for more details).

Value

A list with the following elements:

UP	The upper confidence bound.
ES	The MLEs
LO	The lower confidence bound.

Note

Note that the tcrit has a different interpretation for the Nelder-Mead algorithm and different for BFGS

Author(s)

Vasileios Giagos

References

Giagos, V.: 2010, Inference for auto-regulatory genetic networks using diffusion process approximations, Thesis, Lancaster University, 2010.

See Also

The model parsing is described in parsemod.

Examples

## Not run: 
##We consider the Lotka-Volterra Model
tt <- matrix(c(1,-1,0,0,1,-1),nrow=2,ncol=3,byrow=TRUE)
rfun <- c("con1 * Prey","con2 * Prey * Predator","con3 * Predator")
thetas <- paste("con",1:3,sep="")
species <- c("Prey","Predator")
cout <- parsemod(tt,rfun,thetas,species) #Parse the model

##Inputs a dataset
mydata<-c(0.0, 5000.0, 3000, 1, 5989, 2992, 2, 7165, 3107, 3, 8534, 3306, 
	4, 10041, 3709, 5, 11624, 4265, 6, 13306, 5181, 7, 14741, 6492, 
	8, 15867, 8337, 9, 16025, 10981)
mydata2<-t(matrix(mydata,3,10)) # Put the data into a matrix

compmod(cout,"derivs") #Compile the model

##Test that derivs is working.
derivs(mydata[1],c(mydata[2],mydata[3],c(0,0,0,0,0)),
       rep(0,7),c(.1,.0001,.1)) 

##Run model
run1<-optmod(cout,nthetas=c(.5,.1,.2), mydata=mydata2,
             maxiter=300, simplexsize=1e-5, relerr=1e-9,
             abserr=1e-9, hessianh=1e-4,dfunction=derivs)

##Try a random starting value. WARNING! May crash!
(nthetas<-exp(runif(3,-3,1)))
run2<-optmod(cout,nthetas=nthetas, mydata=mydata2,
             maxiter=500, simplexsize=1e-6, relerr=1e-9,
             abserr=1e-9, hessianh=1e-4,dfunction=derivs)

## End(Not run)

lnar documentation built on May 2, 2019, 4:51 p.m.