As of today (or sometime in the last week), I finally lost access to the school computers I was running the RobocodeLittleLeague battles on. That means that I'm now running the competition on my own home-computer muscle, consisting of a 3-year-old desktop and a zippy new laptop. What this also means is that unless I can have dependable help every Saturday (well, at least as dependable as me, that I can be in semi-regular contact with), the tournament will take a large piece of my time every week :-) (not active time, but computing time if I was going to have any!) Some of this can be alleviated if I can use work computers to hold some of the load (for some reason I couldn't seem to get access to those, either today). However, one thing I should consider is trimming down the divisions I run, or deciding if my tournament is going obselete. I'll probably put up a poll or something (once I decide what it should have on it) either here or on the RLL website. I'll keep you posted.
Some potential ideas to offset the load:
Some potential changes we could make while we're at it:
-- Kawigi
Just a random thought - are you aware of / have you considered the Swiss system, used extensively in the chess world for tournaments with hundreds of players, for which an all play all format would take forever.
The basic idea is that every round, opponents are chosen amongst competitors on the same number of wins, although avoiding pairings that have already happened. Thus (if we assume for simplicity that there are no draws - which is quite achievable in robocode :))) as more rounds are played the original lump of competitors all on zero wins spreads out, and competitors float to their appropriate level in the tournament rankings.
Another feature is that competitors tend to compete more against their peers than against those much stronger or weaker, which would probably produce a slightly different result than the overall Elo-style ratings.
-- Shrubbery
Yes, I am aware of the Swiss system (which was what I was hinting at with improving the pairing engine) - but only because that's what the MiniBotChallenge used. -- Kawigi
My thoughts: 1) I'm willing to run some of the competitions if you want. 2) I would consider the standard division more important than the experimental. But maybe you could go to a system of standard one week, experimental the next. Or experimental once a month or something. 3) Double elimination tournaments are really, really, really cool. -- Alcatraz
Yes, a double elimination system would be nice, or may bee triple elemination. And i offer you too to run some of the battles! --Krabb
How would a double or triple elimination system work for melee, something like places 1 to 6 are counted as victory and 7-10 as loss? Personally I find the standard divisions more interesting than the experimental ones. I also like the 'overall' ranking. If you are low on processing power I offer to run some of the competitions too. Alas I can not guarantee that I am available every weekend. -- GrubbmGait
Anyways, at David's urging, I created a program based on the RoboocodeLittleLeague? automation system that will allow people to run test tournaments using its scoring system. It also includes several useful little utility extensions I made at some point in time, including-
It can run melee or 1-on-1 battles the same way the tournament does and generate html results. The source is included to be hacked away at, so you can see how I create the (exaggerated) rankings that I do in the tournament in both melee and one-on-one. It's available from the RobocodeLittleLeague webpage ( http://robocode.yajags.com/ ) on the downloads page, and more information and examples of how to use it are there. -- Kawigi
I took a bit of time today working on the RLL stuff, and now more of it is automated than ever for me. (People always said this is how to keep an active league even after the person running it loses interest in Robocode). All in all, I just download and delete bots from a couple directories on my computer (well, technically it's a school computer). I can then just type a command and it will make any necessary updates to the lists on the website for me. If you've been on the site at the same time as me in the last two days, you've probably seen test division files get created and deleted on the tournaments page, and several messed up normal pages here and there. That's because now I can also just push the button to synchronize the results on the website with the local results of the tournament as its running. It also automatically generates the tournament index page that displays the winners of each tournament and best overall robots in each division. I regenerated all of those pages, and you'll notice they now display a message saying if they are all-time records for their division (or were at the time of that tournament). Records are also automatically updated and the Records page actually exists now (as opposed to being some kind of 'coming soon' page). The table on the main page that shows the latest results also updates automatically. Next up is the promised promotions between divisions for doing well in tournaments. At the moment I am considering promoting the top 10% of the division, up to 4 total, to compete in the same kind of competition at the next weight up. This would apply to Nanos, Micros and Haikus being promoted temporarily to the Micro, Mini, and Sonnet divisions, respectively. If I have that done by either Friday night or Saturday morning, I'll probably run it that way, but it may take several more hours to run the tournament. Eventually when I feel like everything is stable and things are more automated in general, I'll probably start just running it overnight like I originally intended to. -- Kawigi
I implemented promotions, the rules page has been updated to reflect this. I'll probably make changes to the text of the rules and divisions page and stuff to reflect discussions on this wiki (people here know the rules, even if they aren't down in writing, but they should still be written). Promotions work as follows - I take the top 10% of robots from the nano, micro and haiku divisions (but no more than 4 from each division) and run them in the micro, mini, and sonnet divisions. Melee and 1-on-1 promotions are handled completely seperately.
I found a bug in the code that stores total stats for each competitor. Up until half way through tournament 20031004, the last survivor score was displayed twice instead of displaying the regular survival score. The micro and mini competitions should have correct stats this tournament.
On the whole, the things I've been doing to handle promotions and automate updates on the website have gone smoothly, with a few crinkles to fix here and there. That makes me happy :-) Of course, crinkles make me sad :-( (I know, the smileys help you figure out that I'm sad and happy, you wouldn't have known otherwise). -- Kawigi
Update - 10/18/2003 - I just finished running my smoothest tournament ever. One JVM crash, and everything else worked like clockwork without me touching it. Found a bug that screwed me up when I tried to run divisions consecutively with promotions, which held the previous tournament from going as smoothly. -- Kawigi
Update - 12/1/2003 - I just added an ordering list for the "Overall" rankings. The results server program has been figuring out the best overall bot for some time now, now you can actually see where your bots place in that standing. For instance, now you can tell that Nimrod is not only the best overall mini almost half the time, but that it has been in the top 3 or 4 all except twice. You can also see that not only has FloodSonnet been best overall Sonnet since it was made, it is also doing it by more than 100 points pretty consistently. The calculations I do for this are different from the MiniBotChallenge and don't favor melee specialists as much. In fact, because the deviation of scores is larger in duels, it actually favors one-on-one bots somewhat, unless they are just plain good enough at both (F-Micro isn't winning any overall tournaments until it stops crashing in melee battles). Basically, the top of the overall latter is almost always composed of bots with positive scores in both categories.
![[Home]](/images/RoboWiki.png)