[Repast-developer] FW: Strange interaction between uniform and normal

[Nick C. <nic...@ve...>]

Walter,

I received the following from the colt list. The normal distribution does
indeed draw two from the stream and caches one. So, I think this explains
the behavior.

Nick

-----Original Message-----
From: Adrian Kujaneck Agogino [mailto:ag...@ec...]
Sent: Tuesday, October 19, 2004 2:23 PM
To: Nick Collier
Cc: co...@li...
Subject: Re: Strange interaction between uniform and normal


It looks like the uniform distribution asks the random generator for a
single number for each call to "nextDouble." The normal distribution
asks for two, generates two normally distributed numbers and caches one of
them. If they are drawing from a common random number generator, this
would explain what you are seeing.

-adrian

On Mon, 18 Oct 2004, Nick Collier wrote:

> Sorry, if this is a repeat. It may not have sent the first time.
>
> Hi all,
>
> I've noticed a strange interaction between doing draws from the uniform
> distribution interleaved with drawing from the normal distribution.
>
> The basic scenario is doing some number of int draws from the uniform and
> then doing 10 or so double draws from the normal. It looks like there is a
> strange odd / even kind of behavior w/r to the number of uniform draws.
For
> example, if I do 7 draws from normal, I get:
>
> A, B, C, D, E, F, G
>
> where A etc. are some appropriately distributed random numbers.
>
> If I precede these draws with a SINGLE int draw from a uniform
distribution,
> and then do the normal draws I get:
>
> M, N, O, P, Q, R, S.
>
> That is, I get completely different numbers than in the first case. The
> sequence is different.
>
> However, if I precede the normal draws with two int draws, and then do the
> normal draws, I get:
>
> C, D, E, F, G, ...
>
> The point being that the sequence of normal draws is the same as the first
> case.
>
> If I do three int uniform draws, and then do the normals, the sequence is
> the same as that in the 1 int uniform draw case. I've tested this up to
five
> int draws and the pattern holds.
>
> So, I'm not sure this is a problem, but can anyone shed some light on
this?
>
> thanks,
>
> Nick
>
>
> ----------------------
> Nick Collier
> Software Developer
> Note New Email: nic...@ve...
>
>
>