On Jun 19, 4:41=A0am, gadolin <t...@[EMAIL PROTECTED]
> wrote:
> I am having problem with experimental results.
>
> I am analysing algorithm performance, which has random nature.
> Since its for security application the more random the better, which
> will hide small dependencies under.
>
> What i do is comparison its structure(it results in directed graph with
> each vertex having one outer edge) to random mapping.
>
> Now run tests, and look for graphs with number of vertex 128,
> create random mapping graph, and graph for my algorithm and checks
> number of cycles of different length and so, the number of different
> graph is 128^128, what is my problem know that for number of test
T=3D100=
00,
> the results for both doesn't converge and oscillate.
>
> Could you tell me how many test need to be run to get more convergenous
> results? or point some source where such info can be got.
Are you sure you generate 10000 _independent_ random graphs? Which
particular parts of the results do not converge? Perhaps you looking
at some extreme percentile?
Aniko
|
