Life In 19x19

Mef wrote:
This is precisely the problem...many of them SHOULDN'T HAVE A 50% WIN RATIO.

A 50% win ratio implies that there is no difference between a weak 4d and a strong 4d...there are only average 4ds. It implies that you have no predictive capacity, because the result of every game is, in fact, random chance.

No, you have predictive capacity to tell that this player will likely win against 3d players. The ranks on Tygem are discrete, so there is by definition a range of skill within a particular skill bracket.

Kirby wrote:

Mef wrote:
This is precisely the problem...many of them SHOULDN'T HAVE A 50% WIN RATIO.

A 50% win ratio implies that there is no difference between a weak 4d and a strong 4d...there are only average 4ds. It implies that you have no predictive capacity, because the result of every game is, in fact, random chance.

No, you have predictive capacity to tell that this player will likely win against 3d players. The ranks on Tygem are discrete, so there is by definition a range of skill within a particular skill bracket.

To confirm that would require comparing win rate data in even games between Tygem 3d's and 4d's...

Mef wrote:
To confirm that would require comparing win rate data in even games between Tygem 3d's and 4d's...

My goodness, Mef. I think you're just being silly now. I've given you data from a number of players, and we can see people that have played thousands of games, winning around half of them against evenly matched players.

What more do you want?

If the guy I linked shouldn't be 4d, then what should be be? 3d? Why? He wins half the time against 4d. 5d? Why? He loses half the time against 5d. He's at 4d, and that's just right - he wins half the time and loses half the time.

I'm satisfied with having provided data for people to make their own decision. I encourage everyone to try it out and make their own opinion.

Kirby wrote:

Mef wrote:
To confirm that would require comparing win rate data in even games between Tygem 3d's and 4d's...

My goodness, Mef. I think you're just being silly now. I've given you data from a number of players, and we can see people that have played thousands of games, winning around half of them against evenly matched players.

What more do you want?

If the guy I linked shouldn't be 4d, then what should be be? 3d? Why? He wins half the time against 4d. 5d? Why? He loses half the time against 5d. He's at 4d, and that's just right - he wins half the time and loses half the time.

I think you're just being silly now, so I'm satisfied with having provided data for people to make their own decision.

I encourage everyone to try it out and make their own opinion.

I've told you quite simply what I want: I want Tygem's rating system to make a set of predictions on rated games, and I would like to see it evaluated on how accurate those predictions are. This has been my stance from the very beginning (even in the other thread!) and has not changed. Showing that players have 50% win rates over 8000 games where you do not know who those games were played against means very little. You should expect Tygem's players to have a range of win rates between 33% and 66%, because these are the promotion/demotion bounds. If Tygem's rating system is overwhelmed by random noise, then it makes sense that all players have 50% win rates.

Mef wrote:I've told you quite simply what I want: I want Tygem's rating system to make a set of predictions on rated games, and I would like to see it evaluated on how accurate those predictions are.

Yes, you'd like the system to be KGS's. I understand.

Mef wrote: If Tygem's rating system is overwhelmed by random noise, then it makes sense that all players have 50% win rates.

This is your speculation. Take any player that has had several thousand games, and show me how it is overwhelmed with random noise. I've already shown you such a player, and we could see that his games were against people of his same rank. This is not noise. Like I said earlier, I know that you will not change your opinion. You are a KGS admin, and you like KGS's system better.

And to be honest, in some regards, I enjoy KGS's ranking system as well.

But calling Tygem's ranking system a farce is BS, and I have a good amount of experience playing on the server to know that it's not true. Based on your comments, I'm fairly confident that I have more experience than you. I've also shown you games to show you that we have consistent wins and losses. Since there are so many players there, unlike KGS, so most games are not handicap games. That's why, when someone has a stable rank, we can expect 50% win rates.

In any case, this conversation is not going anywhere, and I believe I've provided evidence that Tygem's system is a viable one, even if some KGS admins think of it as a farce.

I won't comment again in this thread.

skydyr wrote:You never hear anyone complaining that they kept their rank after losing 10 games in a row either, after all.

Oh well, probably because nobody wants to listen. I prefer to go down a rank or two after a losing streak of - say - 5-10 games (at least in the kyu ranges) and try a new start-up at the lower rank (as long as I know it would not take too long to climb up again). It will probably help me to get over my slump quicklier... So I second Oren to make the system a bit more dynamical. Otherwise (and apart from the 2d?-bug) I like the KGS system most (comparing to Tygem, WBaduk, IGS-Pandanet).

As for creating new accounts, I think many do that, because you can easily do it and KGS doesn't deprecate it. There may be other reasons to create a new account. Those players who play tons of games (>20 per day) are clearly the minority. However the original question "Does KGS rank system encourage users to make new accounts" applies in other cases as well. I am one of the players who rarely play - in 99% I just watch tournament events, such as the Dutch championship last weekend. When I am in the mood to play, my idea is a plan to play more often. Then I create an account and play a few games until the rank has stabilized. The point is, that I always use "automatch" to play games (because it's the only way for me to jump over the wall and get a game). Unfortunately the plan will never be realized, and the one day of play remains the only day of play. After a while, the rank of that account drifts to "2d?" (even if you haven't played for more than 180 days), and when I am in the mood again, I don't want to use that "2d?" account, since I feel that I don't deserve to play anyone of that strength without handi stones (it would probably be different, if I were really around 2d). Then I create a new account again (and use the old one for watching). So in a way the buggy KGS ranking system is responsible for users who regularly create new accounts.

Like the "got stuck" problem this is only *one* reason to create new accounts. There may be others as well that I don't know of.

Mef wrote:To summarize -- Do some people find Tygem's ranking system fun? Sure. Does Tygem's ranking system work for people who just want to play? Sure. Does Tygem's system provide significant predictive power in guessing game outcomes? No.

Good sunmmary for Tygem. I'd just add that the predictive power is significantly better in the 1-6d range. In the kyu ranges it is abysmal (18k are usually stronger than 5k).

Kirby wrote:4702 wins and 4358 losses; 4803 wins and 4622 losses; 8045 wins and 8036 losses.
Sure, there are exceptions, but these numbers are pretty close - people are winning similar to the number of times that they are losing.

This probably applies to 5-18k players as well, however you will very often play with (same ranked) players who are actually 10 stones stronger or 10 stones weaker (and these are *no* exceptions). So the pure number of wins/losses has probably little significance.

The big issue for KGS, and why Tygem is better predictive, is the lack of games for higher ranked players. Huge parts of the day go by with no games going on stronger than 4d-5d. These small samples hurt the whole system, its just basic statistics.

The answer to the OP's question is clearly "yes."

If it wasn't, then more people would have just one account. But many have two or more, including some of our fellow L19ers. What is their motive? Who cares, the system encourages people to do it, so they do.

I found the Kirby v mef argument interesting, but irrelevant to the discussion at hand. The question wasn't which is best, the question was does KGS encourage multiple accounts.

wineandgolover wrote:I found the Kirby v mef argument interesting, but irrelevant to the discussion at hand. The question wasn't which is best, the question was does KGS encourage multiple accounts.

It covered interesting ground, but the summary I got from it is 'Mef prefers kgs's accuracy and Kirby prefers tygem's volatility', without any real change of opinion.

I would answer 'yes' as well. However, I think that this is a smaller influence than some other factors. Probably the ability inherent in KGS to make new accounts is the primary reason why people make them.

Javaness2 wrote:I would answer 'yes' as well. However, I think that this is a smaller influence than some other factors. Probably the ability inherent in KGS to make new accounts is the primary reason why people make them.

I think the uncertainty (from the player's point of view) about their current rank and what would be required to reach the next rank might be part of it. The KGS rank graph generates a lot of myths mainly because people don't understand it. Whilst I don't think Tygem's system will tend towards accurately ranking people it's simple for anyone to understand even a child. Which might be a lot of the attraction to it (and to systems like Elo which are equally easy to understand).

wineandgolover wrote:The answer to the OP's question is clearly "yes."

If it wasn't, then more people would have just one account. But many have two or more, including some of our fellow L19ers. What is their motive? Who cares, the system encourages people to do it, so they do...

I think that a more useful question would be: Does KGS rank system encourage well informed users to make new accounts?

In other words, are we contemplating a defect in the system or a defect in the users?

( Full disclosure: Yes, I have more than one account.
I am almost certainly a poorly informed user. )

I don't think that's a more useful question than the one asked. If the system appears to give more volatility to new users (which it does), then people will create new accounts.

There are also more reasons for multiple accounts than concern about rank. I know people with "anti-social" accounts for when they just want to play and not chat to anyone (the lack of a Do Not Disturb tag for KGS being a core reason for this), people with "drunk" accounts for when they can't stop themselves from playing after a few drinks (how well this works seems to vary ) and so on. Both of these are accounts made about rank concerns but not really because of KGS's ranking system.

This has more to do with Tygem than KGS, but to inject a little data into the conversation about volatility of ratings, I ran a short simulation with the following assumptions:

win 50% against equally ranked opponent
win (55,60,65,70)% against opponent with rank-1
win (60,70,80,90)% against opponent with rank-2
promote/demote after winning 70%/30% of previous 20 games
reset win rate counter after promotion or demotion

In general, players spend significant time at the "wrong" rank. The detailed results depend strongly on the assumed win/loss rate for unequal ranks, which is the feedback mechanism for rank convergence:

55% win rate against rank-1 ==> 44% of games played at correct rank
60% win rate against rank-1 ==> 58% of games played at correct rank
65% win rate against rank-1 ==> 66% of games played at correct rank
70% win rate against rank-1 ==> 70% of games played at correct rank

Changing the recent history promotion/demotion counter from 20 games to 30 games has the following effect:

55% win rate against rank-1 ==> 53% of games played at correct rank
60% win rate against rank-1 ==> 70% of games played at correct rank
65% win rate against rank-1 ==> 78% of games played at correct rank
70% win rate against rank-1 ==> 83% of games played at correct rank