En Garde scoring, yet again

Friday, December 14, 2007

We've been living with the new En Garde scoring for several weeks now, and I think that's enough time to evaluate it. I've gotten comments from several people, and everyone pretty much agrees.

It sucks.

I'm sorry, I really thought it would be better than this. But sometimes you can't tell how something will play out until you see it in action.

There's several problems with the new scoring system. First of all, the three games per day limit can sneak up on you. You can't tell if a game you play will count in the scoring or not. This was originally intended as a diversity control: playing lots of games over and over shouldn't add to your score as much as playing many different people. My number one goal in designing a scoring system is to get lots of people to play as many different people as possible, and get new people playing that way.

The biggest problem is with the point loss system. When you lose a game to someone who is ranked lower than you, you lose some of your points. This was intended to provide a little fluidity in the rankings. What has happened instead is, people are afraid to play new players. That's terrible. But I can understand it. There's just too huge a penalty if you lose to a newbie, and the risk is too great.

So, I have to rip all that out, and come up with yet another scoring system.

I am working on a new system that I hope will fix these problems, and provide an accurate ranking of players while encouraging new play. The system is based on Google's PageRank -- the same system Google uses to rank the results when you do a web search, I want to modify and use to rank fencing players. There are many advantages to this. I think it can solve the problems I've listed above. There are disadvantages too. One disadvantage is that it's really complicated, and hard to explain to people. But I have to do something, and so this will hopefully be another small step forward.

I hope to have the new scoring implemented sometime in the coming week. Like all scoring changes, it will be retroactive, and can possible make radical changes in the current rankings.



There is a very simple point system that works well for encouraging diversity while rewarding good play and penalizing bad.

The core idea is to give players a base score (say 1000). For each game that is played against a new player and won, grant some positive value (say, 100 points). Conversely, losing causes a penalty, say 50 points.

In both cases, playing against the same player again causes both the reward and penalty to scale. A second game counts (say) half as much as the first, a third half as much again. A cap can be placed at (again, just throwing numbers around) +10 and -5 for a game.

This type of system has several attributes:

1. It rewards active play. On average players will win roughly half of games played and thus will get a net positive score for participation. The amount of reward can be adjusted by scaling the positive and negative outcomes.

2. It rewards diverse play be allowing the scores to diminish to the point where a new opponent will be ten times as valuable (in the example above) as someone that has been a constant player.

3. It discourages "beating up on the practice dummy" for points, as the most you will get from one alt is around 200 points.

4. It avoids the "playing a newbie is scary" problem by not trying to scale the point loss to the other player's rank. While in a pure skill game like chess or go make good use of ELO style ratings, in a game where even the best player can have a bad round with a newbie, adjusting on player rank is probably ill advised.

Godeke said...
December 14, 2007 at 8:41 AM  

I like the idea of pagerank, or something similar, if you can adapt it well. One thing that you might consider, if using this system, is to extend the scoring lifetime from I think it is 2 weeks to 4 weeks? It would give a better base for the link (win) analysis.

Acru said...
December 14, 2007 at 12:41 PM  

I've heard others mention, and I like the idea of implementing something like the ranking that happens in RL fencing. That is, you have tournaments in which one can earn rank. Everyone starts as U (unrated) and there are some tournaments held for getting to the next couple of ranks, then some for rank C and still others for B or A. This perhaps wouldn't be THE ranking method, but you could have good players have a chance to work up to a rank of A such that even if that player's score drops and world rank drops, he still earned the A. I certainly understand polishing how scoring works first, but I think this could be a nice addition to scoring sometime down the line.

Case Cagney said...
December 14, 2007 at 2:13 PM  

Let's see. What are our objects with the point system?

1. Encourage diversity of play, especially avoiding team members massing points against teammates.
2. Encourage frequency of play without unduly rewarding piling up masses of games.
3. Reward success in play.

The only real problem with the system you are scrapping is that high-ranked players have no incentive to play down very far; they stand to gain at most 5 points and could lose 50ish. Risk/reward is out of whack.

Here's an idea.

1. Start every player with 20 points instead of zero, for reasons which will become evident.
2. Every time you play an opponent you have not previously played you gain points. Your first opponent gets you 12, your second 11, then 10-9-8-7-6-5-4-3, and 2 each for 11th and subsequent opponents.
3. Points for touches are unchanged.
4. The winner of the duel takes 15% of the loser's pre-match points.

Let's put World Champion X rated 200 against Newcomer Y with the initial 20. Assume that our champion has played dozens of opponents.

X wins 5-2: X gains 2 (new opponent) + 5 (touches) + 3 (15%) = 10. Old system 5. Y gains 12 (first opponent) + 2 (touches) -3 (15%) = 9. Old system 2.

X loses 5-4. X gains 2 (opponent) + 4 (touches) - 30 = -24. Old system -46. Y gains 12 (first opponent) + 5 (touches) + 30 (15%) = 47. Old system 55.

X is now risking 24-28 to gain 10, instead of 46-50 to gain 5.

The system feeds points to newcomers to help pay their early losses, and always rewards everyone for playing new foes.

Also, when you fight teammates all the time all the team ever gains is touches -- no new opponent bonuses, and it's zero-sum for the result.

Valentine Janus said...
December 15, 2007 at 6:23 PM  

Here's and idea for a preliminary design structure for fencing tournaments as one means of determining, in their cumulative nature, En Garde rankings in SL :). The following process also does away with the enormous advantage that those who fence very often have versus those who are less able to do so.

Often times in fencing competitions, the preliminary rounds are "Round-Robins" and "The Finals," and/or later rounds, are by direct-elimination method (as, for example, some tennis matches are). In a round-robin round, everyone in the tournament fences *everyone* else. Match-ups in "The Finals" & elimination rounds generally pit the best placer in the preliminaries versus the poorst and so-on, and so-forth. The theory behind this is that you want to eliminate the poorest performers first, and end up with the best, and most evenly matched performers facing one-another at the end. As with tennis, it really makes alot of sense, the larger the competition, and the more [elimination] rounds there are.

Placement for The Finals is based upon performance during the preliminary round(s) and is determined by the most no. of wins. In the event of a tie, placement is determined, first, by total touches *against*. (I have created an example tournament comprised of 6 fencers. The example can be displayed as .jpg graphics.) The theory for this is that it doesn't matter how great you looked on the strip, how nice your uniform was or how many touches you've scored, if you've been stabbed through the heart [237 times] & you're dead!(hehehehe). In the event of a tie there as well (as in the example), *then* it is determined by total touches scored against all your opponents. Naturally, there will probably be a larger number of opponents in a tournament, so "The Finals" (& the elimination round(*s*) ) will usually have a greater number of bouts.

Then, the winner of the tournament will get points as will the other two placers, along with a consolation point for those who participated in The Finals. At the end of a season (or month, or whatever period is commonly agreed to), the "Champion" is determined by the total number of points *in competition*. It's as simple as that! As with individual tournaments / competitions, at the end of the season, in the event of a tie, ranking would be determined, first, by total number of wins (including wins in tournaments where the fencer may have been eliminated in the preliminary round(s) -i.e. round-robins and/or elimation round(s)), next, by total touches *against* and, finally, by total touches scored- all of the preceding based upon the entire season's performance.

Food for thought might also include competitions consisting solely of "round-robins" (one or more rounds). This may be another way to go!

This may all seem unusual, but it is often times how things are done in competitive fencing, and it is often times considered a fair and representative means of determining ranking. In the end, this way, everyone fences everyone else [a fairly similar no. of times]!

Finally, so as to emphasize the importance of tournaments as opposed to the sheer frequency that one fences, there could also be a means of disabling the ranking/point/scorinig outside of tournaments, so as to allow for practice fencing that is more "just-for-the-ummmmmm...fun-of-it!"

Champions in SL could be determined, then, by placement in competitions that pit *everyone* against *everyone else* under common & shared circumstances.

Anyway, just food for thought!

Amelie said...
December 16, 2007 at 10:08 AM  

I apologise for the slightly off-topic nature of this comment. Rifkin convened a meeting in-world on Sunday to try to form a committee for organizing large-scale En Garde! activity. While that meeting did not achieve its immediate goal, discussion among several participants inspired me to write and circulate by notecard two proposals.

Rifkin asked me if I could post them to his Procyon discussion group, but I cannot find it :) and haven't been able to touch base with him. So I created a web site with Google Page Creator (I needed one to host SL stuff anyway) and posted slightly tweaked versions of both proposals.

My site with EG! stuff

One proposal addresses organising an SL-wide league for team competition. The second addresses an umbrella framework to let players earn classifications for their performance in tournaments held by many different organisers.

Neither proposal replaces global ranking, by the way. Your comments to me in-world are invited.

Valentine Janus said...
December 19, 2007 at 11:31 AM  

I just wish there was a ranking system that forced everyone to play each other..Then there wouldn't be certain ones that avoided others in games... Playing everyone should be a requirement if you wished to be ranked.

Richard said...
December 24, 2007 at 7:48 AM  

Post a Comment