Re: [igraph] errors using power.law.fit

[Gábor Csárdi]

Martin, Tamás,

On Tue, Nov 16, 2010 at 11:41 AM, Martin Tomko <address@hidden> wrote:
> Hi guys,
> yes, that is my reading of the paper too. NOw, what is the relation of the
> BFGS method suggested by Gabor and the method they use? Can I apply it?

Apply what? BFGS? It's just another optimization method to find the
optimal alpha. Sure, you can even apply each method supported by
'optim' and then take the best alpha, according to the log-likelihood.
But probably this is not needed, go with 'BGFS', it is supposed to be
robust.

> I do indeed have discrete data (as anyone working with graph degree
> distributions would) and I understand that there is no single perfect power
> law adistribution I could fit to it, but I need an approximation. The alpha
> parameter and possibly an estimate of the goodness of fit (btw, I am
> accessing the result of the power law fiting as follows:
> fitdegS4<-power.law.fit(degrees,method="BFGS")
> alpha<- coef(fitdegS4)[[1]])
>
> is there any simpler method to get the alpha and if possible the parameters?
> I am still not sure how best to access S4 objects, seems like you need a
> method, and I only found coef for a mle -class object, but nothing for the
> goodness of fit parameter which actually gets displayed if I only list
> fitdegS4....

Read ?"mle-class". 'confint' gives you confidence intervals, 'logLik'
the log-likelihood, etc.

Gabor

[...]

-- 
Gabor Csardi <address@hidden>     UNIL DGM