Flattened handicaps

[SmoothOper]

There is no reason to assume I would oppose free placement or alternative komi or time odds or some other such factor. There is also no reason to assume I would want to play hamete.

Though there may be not reason to assume that you would not be opposed. <-edited for clarity!

[Shaddy]

I'm curious- can you give an example of a hamete you've encountered? I have a general interest in hamete (I'm kind of a collector ^^;) since a lot of them tend to be really cute moves that juuust don't work, and sometimes they're strong ways to play when you have extra stones around.

[Tami]

People tend to like to think they`re more accomplished than they really are. That`s one reason they ask for lower handicaps than they need or ask weaker players to take higher handicaps. Probably we`ve all been guilty of it on (many?) occasions. Conversely, people who are genuinely good at something tend to underestimate their ability. How many of us do that?

Before anyone researches it and points it out to me, yes, I have been known to play "enhanced handicaps" games on KGS, and I have my reasons to do so. If the opponent accepts such a match, then that is a fair contract with the risks known on both sides (and I do pretty well usually). But I know that I`m still not quite a dan player, much as I would like to be. Getting the "d" after my handle is not so desirable as knowing that I know what I know.

Anyway, the point is that probably the quickest way to get somewhere is to recognise you`re not there yet. If you need to walk 9 miles to get to Rome, then that`s what you do. If you need to work through 9 stones to reach 1 dan, then that`s what you do. Kidding yourself that you`re in Rome when you`re still in the outskirts isn`t going to show you the Collosseum and the Forum, and pretending you`re an expert in go, and already worthy company for the likes of Artem and Shikshin, when you`ve still got basic things to learn is just as self-defeating.

As for the KGS rating system, the best and most illuminating guide to its mysteries I've ever seen is http://senseis.xmp.net/?HowToGetAlongWithKGSRatingMath. You might get more mileage out of reducing the number of rated games you play than out of attempting to fiddle with the handicaps. I only wish to a very high place that I had seen this article a long, long time ago, because it would have saved me a lot of frustration.

[Loons]

I think the desire to play uphill ranked, underhandicapped is separate from an inflated sense of one's own rank. (Possessing both, as I do, is mere coincidence.)

Ranked games are serious. Uphill games are hard nigh impossible to win. What more could you ask for?

It does have the material benefit that if you became stronger, you would rank faster with these. Conversely, the normal likelihood of you winning is much lower.

[Tami]

My last post wasn`t aimed at anybody, and if it were, I would be as worthy a target as anybody.

Certainly, playing a little bit uphill can only be good for you. You can`t get better at anything unless you make the effort to step up a level, to break out of the comfort zone, and to be stretched to the limit. On the other hand, if the incline is too steep, it might be like trying to push a truck up a cliff - just too hard.

I see using "enhanced handicaps" as a kind of gamble - you get more if you win, but the odds of winning are theoretically reduced. If, and only if, things work out as statistics would predict, this ought not offer any more benefit than playing normally. Now, the reason I do it sometimes is because I rarely win or lose by a narrow margin. Therefore, if I beat somebody weaker than me, I might as well give them a reverse komi of a few points, because that would not affect my winning percentage in any material way, but it would affect the "value" of the win. Of course, I am really aiming at getting stronger, so I won`t have to think about such machinations anymore. Besides, the reverse komi method might not help other players, because they might not have the same tendency to lose either spectacularly or win large.

And, Loons, we really must resume our rivalry soon. You can make me stronger, and maybe I can give you a push too.

[Codexus]

As for the KGS rating system, the best and most illuminating guide to its mysteries I've ever seen is http://senseis.xmp.net/?HowToGetAlongWithKGSRatingMath. You might get more mileage out of reducing the number of rated games you play than out of attempting to fiddle with the handicaps.

This is a well written article. Unfortunately, that's not at all how the KGS ranking system works according to the official explanation.

Players are not promoted based on win ratio. Instead the system uses the historical data to compute the probability that the player is at a certain rank and chooses the most likely rank. In simple words, you get the rank that best explains your results so far.

Now a concrete example. If you're a solid 8k with 50% win/lose ration playing as a 8k. Then you attend a go retreat with professional teachers for a week and you get so many insights that when you come back, you're now playing as a 5k. If you keep playing with a handicap appropriate for a 8k, you're going to win a lot. But since you have many losses as 8k in the bag, 8k is still going best bet of the system to explain your global performance. It's going to take a long time just to get to 7k and then the system will still not have any strong evidence that you're better than 7k.

If on the other hand, you start playing with the handicaps appropriate for a 5k, and start winning half your games. At first your rank won't change much, as your wins will be best explained as a fluke. But if you continue to get consistent results, there will be a tipping point when it will become more probable that you're actually 5k with some 8k losses than the other way around. So your rank will start changing rapidly and will not stop very long at 7k or 6k. (More recent games are considered more relevant by the system.)

So it's my opinion that in order for the KGS system to rank you appropriately, you should play at a handicap appropriate for your real strength. Now the problem as that many people will want to play beyond their abilities which won't do them any good and can be annoying for others but that's an entirely different problem.

[Tami]

This is a well written article. Unfortunately, that's not at all how the KGS ranking system works according to the official explanation.

Thanks for setting me straight. The only thing I`m really convinced of now is that when playing on KGS you`d better think twice about playing straight away after a defeat, because if you go on tilt you`re really going to suffer the consequences. WMS has argued that people also have good days, but who would really disagree that it's much, much easier to go on tilt than to go on a rampage? Not only do we have to improve at go to get a higher rank, but we have to improve our self-control (which probably isn`t such a bad thing to work on).

Anyway, in the future we`ll all be more concerned with our Kaya ranks, so the whole thing is moot.

[Mef]

As for the KGS rating system, the best and most illuminating guide to its mysteries I've ever seen is http://senseis.xmp.net/?HowToGetAlongWithKGSRatingMath. You might get more mileage out of reducing the number of rated games you play than out of attempting to fiddle with the handicaps.

This is a well written article. Unfortunately, that's not at all how the KGS ranking system works according to the official explanation.
Players are not promoted based on win ratio. Instead the system uses the historical data to compute the probability that the player is at a certain rank and chooses the most likely rank. In simple words, you get the rank that best explains your results so far.

At the danger of venturing into the realm of threadjacking...for all practical purposes of someone trying to make sense of their KGS rank, that article is good, and the major takeaway point is sound:

If you play games at a constant rate, the amount of time it takes for your rank on KGS to change will not be affected by the number of games you play.

Getting into the real math behind it with game weighting over time, and most likely ratings, etc is interesting for the mathematicians, but not something someone applies when looking over their game history. Looking at a winning % over the last half-life of your account (15-45 days depending on your rating) gives you a pretty good snapshot of what KGS is aiming for.

Now a concrete example...*snip example*

If I get some time I may try to hunt down the old page on sensei's about this, but if I recall correctly, if you were playing games at a constant rate and instantly improved exactly 1 stone in strength, it would take approximately 1 rating half-life to promote. If you go on a tear and win all your games (what might be expected as a 5k playing at
handicapped at 8k) it would take approximately half a halflife (a quarterlife?)*.

So it's my opinion that in order for the KGS system to rank you appropriately, you should play at a handicap appropriate for your real strength. Now the problem as that many people will want to play beyond their abilities which won't do them any good and can be annoying for others but that's an entirely different problem.

This is sound advice as well, playing games at the appropriate "true" handicap will help sort things out the fastest. It's worth noting though, that this is not necessarily good to do on a non-KGS server (IGS, Kaya, etc), as playing games with a mismatched handicap on those servers will unduly penalize your opponents for helping you correct, and drag their ratings down (it will register all the games as 5k's losing to an 8k...then 5k's losing to a 7k, etc). On those servers if you have a large swing in rank, it would be better to start with a fresh account (or they might have an option to request a rating reset?).




*If memory serves correctly, the example they did on sensei's assumed your rating is confident and the confidence did not drop during either of these streaks. In a real situation, KGS would lose confidence in your rank as your started winning a lot of games, and then it would move faster.

[SmoothOper]

So this is how the IGS ratings work:

http://www.pandanet-igs.com/communities/pandanet/60

I see when you are at the boundary between ranks, there is a delay to get over/under the threshold. Probably like three games.

[oren]

but who would really disagree that it's much, much easier to go on tilt than to go on a rampage? Not only do we have to improve at go to get a higher rank, but we have to improve our self-control (which probably isn`t such a bad thing to work on).

For the sake of argument, I'll go disagree. Let's say I have time on the weekend to play five games. I'm on my top form because I'm not working that day. Then during the weekdays after work I play and then lose those games.

By your argument, I should keep going down in rank even though my rate was 50% wins/losses? Or do you just want to use some AI to strip losing streaks. The problem with that is that the system is made to make games even so if you do this, very often you will have people beating of higher ranks who are just "on tilt" and their ranks not suffering.

I just don't find your argument too practical for a rating system. In fact if you go on tilt in Tygem or some other systems, it's far harder to get back.

[Lamp]

On go servers like KGS, the system defaults to giving you full handicaps--a 3d player gives a 1d two stones (I believe that on either OGS or DGS, the system goes further, adjusting komi up or down to give handicaps that are even more accurate).

In contrast, at the go clubs I've played in, we're more lackadaisical about handicaps. At the beginning of a game, you might talk about who you've played against before, results, whether you took handicap from them. Perhaps you'll mention your KGS rating. If it seems close, and you haven't played before, you'll often hear "let's just play even and see how it goes". Of course these clubs didn't have formal rating systems at all (they were AGA clubs where not everyone necessarily had up to date AGA ratings). If the gap is obvious, or large enough, you'll use a handicap, but otherwise you might not sweat it. I think some tournaments handle handicaps by reducing all handicaps given by one or two stones.

The more I think about it, the less sensible the first system seems. What goal is served by trying to handicap every game to exactly 50-50? In other games I've played, you just play without a handicap. Those games don't have handicaps, so they had no choice, but the sky doesn't fall. If a game is too unbalanced that does suck, but I never catch myself thinking "this player is a 4:1 favorite against me, I better just give up." As long as the odds aren't miniscule, you will eventually win, and it will stick with you.

I'm wondering what the tradeoffs are for a system when choosing between full handicaps and diminished handicaps. It seems to me that one positive influence is that without handicaps, a quickly improving player should be able to move his rank more when playing against stronger players. Conversely, an overrated player might stay stuck at a high rank longer. Also, rank obsessed players might become more averse to playing weaker players. Lastly, it's possible that it would do bad things to the ratings math--though I don't know if there's any reason to actually be worried.

What do you think?

The entire basis of the rating system is based on the handicap system. The definition of a strength difference of "2" (e.g. 6k vs 8k, 7d vs 5d) is that with a 2-stone handicap you will win 50% of the time.

So sure, you can play with more relaxed handicap rules and there's nothing wrong with that, but just remember that ratings would not be able to exist in any kind of meaningful fashion without, or at least not with their current meaning, otherwise.

[hyperpape]

The entire basis of the rating system is based on the handicap system. The definition of a strength difference of "2" (e.g. 6k vs 8k, 7d vs 5d) is that with a 2-stone handicap you will win 50% of the time.

So sure, you can play with more relaxed handicap rules and there's nothing wrong with that, but just remember that ratings would not be able to exist in any kind of meaningful fashion without, or at least not with their current meaning, otherwise.

Are you sure about that? Have you proven it?

[jts]

The entire basis of the rating system is based on the handicap system. The definition of a strength difference of "2" (e.g. 6k vs 8k, 7d vs 5d) is that with a 2-stone handicap you will win 50% of the time.

So sure, you can play with more relaxed handicap rules and there's nothing wrong with that, but just remember that ratings would not be able to exist in any kind of meaningful fashion without, or at least not with their current meaning, otherwise.

Are you sure about that? Have you proven it?

Proven what... isn't that just true? ("that"=In an accurate rating system, a properly handicapped game gives a 50% chance to each player, and this is all that the ranks mean.)

[Lamp]

The entire basis of the rating system is based on the handicap system. The definition of a strength difference of "2" (e.g. 6k vs 8k, 7d vs 5d) is that with a 2-stone handicap you will win 50% of the time.

So sure, you can play with more relaxed handicap rules and there's nothing wrong with that, but just remember that ratings would not be able to exist in any kind of meaningful fashion without, or at least not with their current meaning, otherwise.

Are you sure about that? Have you proven it?

Not sure what you mean, prove what? It's the definition of the rating system, there's no need to prove definitions.

I'm sure you could construct a rating system around a different method of handicapping, but it would be just that: a different rating system.

[SmoothOper]

The entire basis of the rating system is based on the handicap system. The definition of a strength difference of "2" (e.g. 6k vs 8k, 7d vs 5d) is that with a 2-stone handicap you will win 50% of the time.

So sure, you can play with more relaxed handicap rules and there's nothing wrong with that, but just remember that ratings would not be able to exist in any kind of meaningful fashion without, or at least not with their current meaning, otherwise.

In IGS the rating system is technically different than the ranking system, and the ranking system is based on the rating. IE you could be a weak 6 kyu playing a strong 8+ kyu and there be less than two stones difference in relative strength, as the rating go. IE most the time if the 6 kyu gives the 8+ kyu two stones he will lose.