the random number problem

[jtollison78]

I remember seeing someone comment(Prock maybe?) on there being several mechanical difficulties to producing useful simulations including something called the random number problem. This stuck out as I wondered if it had anything to do with a small random number seed that led to the site being cracked years ago.

If anyone could comment on the random number problem as it relates to simulations and/or any other problem I might run into while trying to write my own simulations, I would be thankful.

If my memory is doing me a greater diservice than normal, and this makes no sense, please ignore me.

Thanks,
John

[Andrew Prock]

I'm not sure which problem that you're talking about. But the problems that I've encountered have to do with selecting scenarios for monte carlo evaluation. One pseudo random number is easy to generate. Take the problem of selecting cards at random.

If you select two cards purely at random, there is a chance you will select the same card. If you have many hands to fill out, and you pick purely random cards, then the more cards you select, the greater the chance of a collision. This is why the old version of PokerStove would go very slow with lots of random hands.

The problem is that if you select cards for a specific hand first, and remove those cards from the deck, then you're not generating scenarios uniformly over all possible scenarios. For monte carlo to work, the scanrios generated must be drawn uniformly from all possible scenarios.

Solving this problem efficiently is non-trivial.

I hope that helps.

- Andrew

[A.Nironen]

[ QUOTE ]
The problem is that if you select cards for a specific hand first, and remove those cards from the deck, then you're not generating scenarios uniformly over all possible scenarios. For monte carlo to work, the scanrios generated must be drawn uniformly from all possible scenarios.

Solving this problem efficiently is non-trivial.


[/ QUOTE ]

The problem is solved long ago. Google for Fisher-Yates shuffle algorithm. (I invented it indepedently [img]/images/graemlins/smile.gif[/img] )

[Andrew Prock]

woot!

tx for the ref

[img]/images/graemlins/smile.gif[/img]

- Andrew

[Lottery Larry]

I didn't understand this:

[ QUOTE ]
The problem is that if you select cards for a specific hand first, and remove those cards from the deck, then you're not generating scenarios uniformly over all possible scenarios. For monte carlo to work, the scanrios generated must be drawn uniformly from all possible scenarios.


[/ QUOTE ]

Isn't it the POINT of selecting specific hands to get the scenarios for a pre-existing condition that affects the other cards? All you are really doing is filtering the entire set of simulations by choosing only those with XY in a specific hand.

Why would that make any real difference in the results?