Posts:
1,031
Registered:
Dec 16, 2001
From:
Powell, OH
Age:
37 Home page

Attention all turn-based players! Sets are here!
Posted:
May 3, 2006, 11:21 PM

All turn-based rated games must now be played as 2 game sets, you might have heard about sets before or maybe not.

In most games at DSG player one has an advantage. There have been various tweaks and variations proposed to the games to attempt to minimize the advantage. However, instead of changing the game rules, you can remove the player advantage by simply requiring that 2 games be played, with players switching sides in the 2nd game. If one player wins both games then the set is a win for that player, and ratings for each player are adjusted. If the result of the set is one win for each player, then the set is a draw, and no rating change is made.

So, now when you start a turn-based game, if the game is rated the set of two games will be created automatically for you. Both games in the set are played at the same time. If you play an unrated game you can still play just a single game and choose p1 or p2.

For those games that have already been played or are being played I have updated them to be unrated. I have also made the difficult decision of wiping out all turn-based statistics. I know, it sucks but no one had really built up that much of a record yet and it is the cleanest way to do it. I was #1 in Turn-based Keryo-Pente so its not like I'm gaining anything by doing this!

Also note that you will remain provisional in a game until you complete 20 sets that are not draws, not just 20 games like in the old system.

Turn-based games are coming along, hopefully I can find some more time soon to complete a few more things and then I will release it to the whole world. For now I'm going to allow some more players in to try it out, if you know someone you'd like to play in turn-based who isn't already a donor you can email me at dweebo@pente.org and I'll let them in. I'm going to let in 50 new players so act quickly!

Let me know if you have any problems, questions or concerns. Set-based games are a very big change, hopefully in time everyone will get used to it.

Finally I'd like to thank up2ng for his tireless promotion of sets, he's been arguing for this for years so you can thank him or scream at him about it

Re: Attention all turn-based players! Sets are here!
Posted:
May 11, 2006, 9:47 AM

Ok, posting for ever how long gets the cookie, right?. Congrats to Dean for that. That doesn't rectify the problem if it doesn't rectify the problem. yeah right, i said that.

Let me promote an example of the fallacy of set based ratings, as if it weren't obvious. A 2000 level player vs a 1200 player; Each player wins as player one, or better yet each wins as player 2. In my earnest opinion the 1200 player should ADVANCE! while the 2000 level player should FALL FREAKING DOWN ON HIS BUTT!! As it is in a set based rating system, each players rating doesn?t change at all.

A weighted rating system on individual games is THE ONLY way to properly apply rating changes to a given game. set based ratings r totally unfair in many situations where a lower rated player splits any set with a high rated player, and comes out with ZIPPO. A lower rated player splitting a set with a high rated player should advance every time, and if that same low rated player wins as player 2 he should advance even more.

Weighted ratings in each individual game is the only way to go in order to be fair to everyone, not just the high rated player.

As to what weight in rating changes should be applied to any given game vs whoever,? well provide me an excel file with all games in the db, profiling wins vs losses, and rating differences in each of those games, and i will provide u with a real weight value to apply to each game based on the ratings difference, and on a player 2 win as opposed to a player 1 win.

The set based equation, though an improvement (thanks Dean), is not AT ALL the way to go. Period!

Ok Peter, I know u r experimenting with a new concept, and I can?t fault u 4 that, but please consider the reality of my points before u make a final decision on the set based ratings issue. I am glad to c u r testing them out on turn based 1st rather than implementing them across the board.

Posts:
542
Registered:
May 9, 2002
From:
Northeast USA
Age:
38

Re: Attention all turn-based players! Sets are here!
Posted:
May 18, 2006, 11:36 PM

Richard, you need a beatdown.

Dweebo, you're absolutely right, this objection has already been raised in previous threads and has already been shot down with authority.

Richard, your solution is basically equivalent to the current system. We already have a weighted ratings adjustment system based on individual games and although it is better than nothing it has already been proven to be flawed at its core. The particular point of the set based approach that you are objecting to makes me realize that you are not grasping some basic principles, which is surprising because I know that you are very good at math.

Unfortunately I don't have the time or energy to elaborate, but I will when I find some free time. Or you can just find the previous thread where all the math is layed out.

Re: Attention all turn-based players! Sets are here!
Posted:
May 19, 2006, 9:34 PM

EL WRongO!

I work with weighted solutions every day in my line of work, and understand the math perfectly well, thank u very much.

The present system only weights ratings changes according to ratings differentials, not with weights assigned to P2 vs P1 wins and loses.

A P2 vs P1 weighted system would result in a P2 win having far greater weight in its effect upon ones rating than a P1 win would have. The reverse would be true for loses.

Let me create a fictitious example: lets imagine if the follow percentages were true.

Speculative average win\loss among near equally rated players: P1 wins 80% P2 wins 20% This means P2 wins 1 in every 5 games vs an equally rated player, or 20% of the time. This translates into meaning that a P2 win among equally rated players should have 5 times the value of a P1 win, and a P2 lose should have 20% the effect upon the loser?s ratings as that of a P1 lose would have.

Exact Win\Loss percentages could be determined by filtering the DB to gather only games among players with less than say 100 points in ratings, and compare those to determine an over all win\loss differential for P1 and P2.

The combination of P2\P1 win\loss differential applied weights in combination with the already existing ratings differential algorithm would achieve dramatic results in how ratings are affected by any give win or loss, and would go a long way towards properly reflecting the real meaning of a win or loss in any give game.

Why is this so hard to understand? Perhaps I just can?t explain it well.

Posts:
542
Registered:
May 9, 2002
From:
Northeast USA
Age:
38

Re: Attention all turn-based players! Sets are here!
Posted:
May 20, 2006, 3:26 AM

Ok, first, I didn't mean to sound harsh in my previous post either. In fact, I was probably going for funny, but I wasn't in a great mood at the time so it probably sounded worse than it should have been. Perhaps I'll go back and edit the post later.

Anyway, first let me show you why both your logic and your math are not correct. Then, instead of beating the correct math and logic like a dead horse, I'll actually use common sense.

"I work with weighted solutions every day in my line of work, and understand the math perfectly well, thank u very much."

Ok, this may actually explain a lot. I now believe that this is precisely your problem with this particular debate. Since you work all the time with weighted solutions, you automatically lean towards finding a weighted solution to any given problem. It's like a patient complaining about an @ss ache. If they see a surgeon, they will be told that they need surgery. If they see a chiaropractor, it's a back problem. A nutritionist will give a nutritional solution and a psychologist will tell you it's all in your mind. In this particular case, although a weighted solution may provide some improvement, it is not the "best" solution, since a much more elegant and exact solution to the problem already exists.

"The present system only weights ratings changes according to ratings differentials, not with weights assigned to P2 vs P1 wins and loses."

You are actually on the right track here. I have already discussed in previous threads that the current methodology (at least conceptually) uses an equation with one variable to attempt to model changes in a system which contains two unknowns. Your statement above sort of addresses this fact. We know that one of the variables is the ratings (skill) difference, but you are missing the other. The other real factor that plays a role in who wins a pente game or match really has nothing to do with P2 wins vs P1 wins at its core. The factor is the P1 advantage that is actually experienced in the game (or match) and how much influence it has in who wins. Maybe the difference is subtle, and maybe we're saying the same thing, I'm not sure. But, at best, you are trying to model this factor's importance indirectly with approximations and analysis of previous data. The problem is that the solution will be extremely complex and its accuracy will still only be marginal. Here's why. Based on the inherant properties built into a game of pente, the P1 advantage has a specific amount of influence on who wins the game. In fact, this amount is infinity, reducing the factor of ratings differences to 0. In other words, in the purest mathematical sense of the game, Player 1 always wins. Always. In practice, this is never the case. Even the best player in the world will lose as P1 in a timed game against a decent player about 1% of the time. As we look at a less and less skilled player, they will lose as P1 a larger and larger percentage of the time against an opponent with the SAME rating differential. In other words, the EFFECT of the P1 advantage actually changes with the skill of the player. So, when 2200 plays against 2100, P1 will win more often than in 1200 vs. 1100. Yet, their differential is the same. The solution gets extremely complex, and it is still an approximation based on the sample size and outliers that exist in our database.

ISN'T IT BETTER TO JUST ELIMINATE THE VARIABLE???? By playing a set and evaluating the result of the set, we know for certain that a win is based strictly on skill and has nothing to do with P1 advantage. A draw simply means that no one was able to sufficiently prove that they were more skilled during that contest -- perhaps the higher rated player was slightly more skilled, but not skilled enough to overcome the significant influence of the P1 advantage. Perhaps not. But why should he be penalized if he played the better game and it just happened to end in a draw? It makes no sense. A draw simply means nothing -- it really means that a winner has not been determined YET, and the players may continue the battle now or at a later date by playing another set.

Ok, now for my attempt at a common sense analogy:

Consider a tennis match. We all know that there is an inherant advantage built into the concept of a POINT in tennis. For those who don't know tennis termonology, a "point" is one rally -- a player serves the ball, there's a rally, and someone wins the point. Perhaps it can be argued that this advantage, in practice, is small with two novices on the court. But it is quite large when two highly skilled professionals take the court.

Consider a match between Roger Federer and Andy Roddick. Although they are both experts, Federer is currently considered the better player and is slightly higher ranked (rated). This isn't just heresay -- they have played against each other several times, and Federer has won the majority of those contests, especially recently. This is despite the fact that Roddick has one of the most devastating serves the game has ever seen.

Now, imagine what would happen if they got together to play a match, which often consists of, lets say, 200 points. But, before they started the match, they flipped a coin to see who would serve the ball EVERY POINT FOR THE ENTIRE MATCH!

In my view, this is what happens in a single pente game between two highly skilled players. If Roddick wins the coin flip, even though he is lower rated, he would destroy Federer nearly 100% of the time. Granted, he could get deathly ill one day and lose one of these matches, but you get the idea. So, after this match, would it make ANY sense at ALL for Roddick to overtake Federer in the world rankings??? No, it would not. No matter how you slice it.

One of the reasons tennis is so competitive and fun is that the scoring structure seeks to eliminate the server advantage by allowing the players to take turns serving. Thus, whoever wins the match was generally the better player on the day -- it has nothing to do with who the server was.

I hope this helps explain my point of view. I consider it to be fact, but perhaps we just have a difference of opinion, who knows?

Re: Attention all turn-based players! Sets are here!
Posted:
May 21, 2006, 8:22 AM

LOL! Ok, This is my last bit of my input on the matter. With no disrespect to my punk friend Dean, i shall totally disregard the previous book i just saw posted. hehe.

Here is a simple weighted solution to the ratings equation. Don't replace the existing formula with a new one, simply apply a weighting to it's end results.

Example: Ok lets assume that my 80%-20% statistic is accurate for the sake of argument.

Bob and Lee have just finished a game. Lets say Bob won and Lee lost and Bob's rating changed +8 to Lee's -4 under the standard ratings adjustment. (the actual ratings for each player is irrelevant)

Existing Standard Result Example: Winner Bob +8 Loser Lee -4

Weighted adjustment applied to existing equation?s results with winner as P1: Winner Bob (+8 * 0.2) * 2.23607 = +3.577712 (rounded to +4) Loser Lee (-4 * 0.2) * 2.23607 = -1.788856 (rounded to -2)

Weighted adjustment applied to existing equation?s results with winner as P2: Winner Bob (+8 * 5 ) / 2.23607 = +17.888? (rounded to +18) Loser Lee (-4 * 5 ) / 2.23607 = -8.944? (rounded to ?9)

x(squared) / 25 = .2 (or 20%) x = 2.23607

This is a differential equation specifically designed to maintain a 5:1 ratio between winning as P1 vs winning as P2, or losing as P1 vs losing as P2. The equation is applied to the results from original standard ratings change. If the actual win\loss ratio for P1 vs P2 is something other than a 5:1 then the equation can be altered to fit whatever ratio actually exists. I thought a post application would be easier to apply to existing results rather than revamping the original equation. You can still set a max change limit to 30 points if u like.

These ratings changes reflect real game value differences of wining or losing as P1 and as P2, that is if the differential ratio is 5:1.