Double Elimination

[BeligerAnt]

Dave, until you've tried it, you just wouldn't believe how complicated it is!
Trying to keep everything (pseudo-)random whilst preventing people fighting themselves as long as possible and allowing for any number of robots makes for some "interesting" logic.

The manual system you suggest basically involves re-drawing part of the competition at some stage. Since it would be a rather complex draw it could take some time and is of course error-prone. At least the fixed algorithm we have at the moment only requires one draw at the beginning then everything runs on its pre-determined path. We just need to tweak the path a bit.

It's fair to say that Oliver,James and I have got the initial draw down to a fine art now

I'll see what I can do to fix the repeating fights. I think we can reduce the likelyhood bya factor of 2 or more quite simply, but I need to work through all of Peter's charts. Currently there is about a 1 in 4 chance of a fight repeating which isn't really acceptable.

[Remote-Controlled Dave]

Dave, until you've tried it, you just wouldn't believe how complicated it is

I have run it before, using Pete's charts at my last AWS, but with about half the robots we seem to be getting now. haha.
I don't think eliminating repeat fights completely is a necessity, just to reduce the likelihood as you say Gary. It seemed to happen a lot at the last AWS. I had 2 robots in competition and both had repeat fights, and Hib having three fights with the same robot seemed a bit daft.

But yes, I am up for a smoothly run event over a pain in the *** one, specially as I am at work at 6am the next morning! haha

[BeligerAnt]

I now have the latest charts from Mr W and will see what I can do...

[Remote-Controlled Dave]

Thank you, sir

[BeligerAnt]

Well, after some mathematical and logical head-scratching I have an improved double-elimination formula:

http://homepage.ntlworld.com/g0xan/doub ... nation.pdf

The charts cover 5-way up to 9-way groups. Larger groups are generated by combining the smaller groups, and moving robots between sub-groups as necessary. There are extra fights to add on; I will generate the complete charts when I have time.

These charts can cover groups from 5 to 16 robots, and Peter has charts for finals based on 2, 4 or 8 groups.

So the charts can cope with any number of entries between 10 and 128 robots. I'm not sure any of us could though!

In the 5-way group there is still a good chance that two robots will meet twice, but the group is too small to do anything about it. The only real solution is to try to avoid 5-way groups. (And 10- and 11-way.)

In all other groups, the chances of a fight repeating are about 1 in 4 (worst case) for a robot that loses its first fight then manages to stay in the losers' stream.

Due to the mathematics of knock-out competitions, only 8-way groups can provide a truly fair solution. All other group sizes mean that some robots will get byes and therefore need to win fewer fights to get to the finals. For this reason robots must be randomly allocated within each group. It's something that James, Oliver and I have got quite good at

Please take a look at the charts and comment, especially if you find any flaws in the logic!

[Remote-Controlled Dave]

That link just takes me to a blank page Gary.

It all sounds very good though...lol

[BeligerAnt]

Hmm works for me (prbably something to do with caching)... will try a different computer and fix the problem tomorrow...

[Remote-Controlled Dave]

My comp is a little tempremental sometimes....is that spelt right? lol

[BeligerAnt]

Dave, I think it's a problem with your computer. I've tested it with Win 7, Win XP, Linux, iOS with IE8, Firefox, Opera, Google Chrome, Safari and Acrobat, Foxit and whatever Apple use to read PDFs!

The most likely problem is that the link between your browser and PDF viewer is broken. Try downloading the file to your PC then opening it locally. The solution is probably to un-install and re-install your PDF reader.

For anyone that can't be bothered to go back to page 1, here's the link again:
http://homepage.ntlworld.com/g0xan/doub ... nation.pdf

[Remote-Controlled Dave]

Yeah, it works when I download it Gary, thanks. My comp has always been a bit odd about PDFs.
It's looking good, I've followed it through and seems ok. Think I'll print a few off and do a test draw/run through, check that it works, and also that I understand it, haha.