At 05:00 AM 29/09/07, Rob Kennedy wrote:
Wilfried Mestdagh wrote:
> Convert it to integers will speed up a lot.
There are 20^24 permutations he wants to analyze. If he can process 1
billion permutations per second, it will take 531 billion years to look
through all of them. Let's not worry about the data type.
Hi All,
thanks for the responses to date. I suspected I might be barking
entirely up the wrong tree and it seems as though I may have
suspected correctly.
Ok, here's what I'm up to. I am using pair-wise matching to predict
the winner of a contest. There are currently 24 contestants, and I
use 20 separate data items for each contestant. Currently I simply
go through all the permutations of pairs and add up the number of
times each contestant wins the pair-off, then convert these to
probabilities. This is working reasonably well.
So, I was thinking why not extend this to attempt to predict
finishing order. This shouldn't be too much harder, or so I
thought. My idea was rather than compare them pair by pair and take
the "winner", I could line up each permutation of data items and take
the first 5 in order, add up the number of times they finish first,
second, third and so on, and convert these to probabilities.
So, the real question is whether there is a faster way than 531
billion years to achieve this. In fact, I'd like to be able to
complete it in a manner of seconds. The current pair-match runs in
under a second.
Cheers,
Chris.
_______________________________________________
Delphi mailing list -> [email protected]
http://www.elists.org/mailman/listinfo/delphi
