fitting: Fitting laws with maximum likelihood

In brainwaver: Basic wavelet analysis of multivariate time series with a visualisation and parametrisation using graph theory.

Description

Fits three laws (exponential, power and truncated power laws) to an empirical distribution using the maximum likelihood estimators.

Usage

fitting(degree.dist, nmax)

Arguments

degree.dist	vector of the distribution to be fitted.
nmax	maximum of the value of degree.dist.

Details

The fitted laws are : exponential law, power law and truncated power law.

This function plots the histogram of degree.dist (dist.ps). This function plots also the cumulative distributions of the empirical and three fitted laws in a log-log scale (fitting.ps). Finally, all the parameters are exported to the file fitting.txt.

Value

mu	parameter of the exponential law.
gamma	parameter of the power law.
alpha, beta	parameter of the truncated power law.
AIC	vector containing the Akaike's criterion for the fitting of the three laws.

Author(s)

S. Achard

References

Akaike, H. (1974) A new look at the statistical model identification IEEE Transactions on Automatic Control Vol. 19, N. 6,pages 716-723.

Examples

data(brain)

brain<-as.matrix(brain)

# WARNING : To process only the first five regions
brain<-brain[,1:5]



n.regions<-dim(brain)[2]

#Construction of the correlation matrices for each level of the wavelet decomposition
wave.cor.list<-const.cor.list(brain, method = "modwt" ,wf = "la8", n.levels = 6, 
                               boundary = "periodic", p.corr = 0.975)

#Construction of the adjacency matrices associated to each level of the wavelet decomposition
wave.adj.list<-const.adj.list(wave.cor.list, sup = 0.44, proc.length=dim(brain)[1])

# For scale 4
degree.dist<-rowSums(wave.adj.list[[4]])

par(cex=1.5,cex.lab=1.2,font.lab=2)
hist(degree.dist,xlab="Degree", ylab="Number of regions",main=NULL,col=1,border=8)

nmax<-50
tmp<-hist(degree.dist,breaks=c(0:nmax))
cum.dist<-1-cumsum(tmp$counts)/n.regions 
# cumulative distribution of degree.dist


d<-fitting(degree.dist,nmax)

exp.trace<-exp(-d$mu*(0:nmax))
# cumulative distribution of the exponential law

power.trace<-(1:(nmax+1))^(-d$gamma+1)
# cumulative distribution of the power law

gamma.trace<-1-pgamma((0:nmax),shape=d$alpha,scale=d$beta) 
# cumulative distribution of the truncated power law

par(cex=1.5,cex.lab=1.2,font.lab=2)

plot(log(1:(nmax)),log(cum.dist),pch=3,xlab="log(k)",ylab="log(cumulative distribution)")
lines(log(1:(nmax+1)),log(exp.trace),lty=3,lwd=2)
lines(log(1:(nmax+1)),log(power.trace),lty=2,lwd=2)
lines(log(1:(nmax+1)),log(gamma.trace),lty=1,lwd=2)
#text(c(0.5,0.5,0.5,0.5),c(-3,-3.5,-4,-4.5),labels=c("+ data","-- power law",
#        ".. exponential law","- truncated power law"),pos=4)

brainwaver documentation built on May 2, 2019, 10:23 a.m.