[jgrapht-developers] RandomGraphGenerator behavior

[Charles F. <cf...@us...>]

It seems to me that the RandomGraphGenerator specification is incorrect.
It currently generates the same graph over and over on subsequent calls
to generateGraph(), and I argue that it should rather produce (probably)
different graphs on each call to generateGraph(). I don't recall any
discussion about the class, but it may have occured before I joined the
group. In any case, I need to extend RandomGraphGenerator, and I would
very much like to see its specification improved. :-)

The closest parallels that I can find in Java itself are Random and
SecureRandom, which generate random numbers as opposed to random graphs.
Their model is to construct a single instance of Random, and then to
generate a (probably) different random number on each subsequent method
invocation. It is possible to reseed the random number generator (or to
seed the constructor) to obtain duplicible results.

I argue that RandomGraphGenerator should do the same. It should be
possible to construct an instance either with or without specifying a
seed, and then to reseed the instance at will. But between reseedings,
succesive calls to generateGraph should create (probably) different
graphs.

I also propose that RandomGraphGenerator be turned into an interface
(following the specification outlined above, which I would be glad to
turn into code), which could then be implemented by the current
generator, as well as the currently experimental
UniformRandomGraphGenerator and PartiteRandomGraphGenerator (shouldn't
that be Bipartite?), in addition to some new RandomGraphGenerators that
I plan to write (RegularRandomGraphGenerator and
MOutRandomGraphGenerator to begin with, though
PreferentialAttachmentRandomGraphGenerator would be nice, if not
verbosely named).

Charles

-- 
My job is
Keeping faces clean
And nobody knows
De stubble
I've seen
Burma-Shave
http://burma-shave.org/jingles/1950/my_job_is