...but I'm sure it's not supposed to work like this.

To start off, it had been a while since I logged onto KGS. My rank when I got back on the other day was set as "15k?" since it had been so long since I'd played a game.

Today, around noon, I played a game with a 14k. He lost a critical fight, ceding me pretty much the entire top of the board, in addition to the lower third I already had. He seemed to want to play out the game, so we did and I won by 133.5 points, pretty much due to that critical fight and not anything special I did right.

"Holy ****," the KGS ranking algorithm remarked to itself. "This guy just won by 133.5 points. He must be really good!" Thus I became "8k?", probably since the ? flag makes it easier for my rank to change until I build up more recent games.

This was a little annoying, because it was going to mess up my automatch options. "What the heck, though," I thought. "I'll just automatch a few games, lose badly in handicapless games against some SDK, and revert back to a more reasonable rank in short order.

So about an hour and a half ago I hit that trusty automatch button, and got pulled into a game with a 9k. He played well, and had me against the wall for most of the game. When I realized this, I was set up to lose by approx. 30 or so points. I figured, since at this point it was so close to the end, that I too would play out the game, lose by 30ish points, and get my rank back up into double digits.

But that was not what destiny had in store. My opponent made two silly mistakes in a row which allowed me to capture one of his larger string of stone and rescue a previously dead group. I ended up winning by 9.5 points, which would be a reasonable margin to win by if I was actually an 8k.

But the KGS algorithm has much more faith in me than I do myself. "Winning by 10 points? Oh my god. That is the point margin of a man destined to have his soul trapped in a goban to help create the perfect game of go!" And thus did I become a "3k?".

tl;dr Through what was clearly diligence and fighting spirit I gained 12 stones in 8 hours.

How much you win by has no effect. If you do not have many games played in the past 3-6 months your account's rank will be very uncertain and will jump wildly around until the system figures out roughly how strong you are based off of who you've beaten/lost to and who they've beaten/lost to. Play 20-30 games and your rank will solidify quite a bit.

there are too many variables at play here..
maybe the 9k player isn't really 9k after all.
maybe you really are better than you think you are.

there are not enough games against reasonably stable players to really tell.

But also, I found that I actually play much, much better against stronger players. They make predictable moves that I usually know how to answer. But when I play weaker players, like myself, they play weird and stupid moves that I don't know how to answer, so I respond with even dumber moves. It keeps me down.

Play even (or less handicap) games against stronger players. Its better for you anyway.

Your level has not changed in 8 hours.

The way the KGS ranking system works, the conversation would have been something like this:

"I wonder how strong this person is....they haven't played in months! Well, let's see how their old opponents have been doing...Oh, look at that, most of the opponents haven't been playing in months either! I guess I'll just keep the old rank estimate, but now I'm going to stick a question mark on it."

*game against 14k*

"Oh, look, a new data point. This person has beaten a 14k, and in their only other game that still seems meaningful, they also won handicapped as a 14k. This person must at least be 14k...but how strong might they be? Well, I know they lost games a long time ago...but I don't really know how strong those people they lost to were. Also a lot of things can change in 5 months...I guess if I were going to pick an upper bound for my guess it could be as high as 2k? For now, I'll split the difference and call it 8k, it's not like this is permanent so once I find out more I can quickly fix it."

*game against 9k*

"There we are another data point! Well, now we can be pretty sure that at least 7-8k isn't an unreasonable guess for a lower limit...This person is still effectively undefeated though, maybe 2k was a little low for an upper limit. Let's slide it to 1d just to be safe...all right, [3k?] it is, we can always fix this stuff later if it's off"

Once you play someone (like a 3k) and lose the system will say:

"Aha, now we have good bound, this person is probably at least an 8k on the low end, but probably not stronger than a 4k on the upper end...let's call them 6k? until we know more."


And then once you play a lot it will be more like

"This person has won X% of games against opponents averaging a strength Y weighted over timespan Z -- This person will most likely have a <some percentage> chance of winning against an average player of rating Q assuming they adjust with handicap H and komi K....therefore I shall rate them as [rank]"

That makes sense. My rank "solidified" at 6k. Now it's at 9k and dropping. Alas, almost depressing.



But I thought I was a prodigy ;_;

Oh! Good luck!

SO how many games do youhave to play before you lose the question mark after your rank?

Until you are willing to make a $100 bet that if you tell me your game results I will be able to deduce your rank to be the same as you think it should be based on those results.

@mef (or any other knowledgable person)

let's say that you lost to a 10k, beat a 1d, lost to a 7k, beat a 2k, lost to a 14k, and beat a 3d. (I know that you should be tossed as a sandbagger but just for the sake of discussion.) Would the algorithm make any kind of 'sense' out of this or would it be left waiting for something consistent?

_________________
Dave Sigaty
"Short-lived are both the praiser and the praised, and rememberer and the remembered..."
- Marcus Aurelius; Meditations, VIII 21

There won't be much difference between that and the "normal" results. If you lose to a 30k and beat a 1d it will just split the difference and put you at 15k.

On a practical note, you would be expected to drift up slowly after these games. Realistically because the 30k will improve quickly, but also theoretically because games for weaker players decay much faster.

I would say that it would immediately bring into question the validity of those opponent's ranks.

If your KGS handle is 'Akashe', then the ranking system is not just working well, it is working brilliantly! It has discovered you are 11kyu, just pushing towards 10kyu, and yet it hasn't even had 20 games yet. I would have expected it to need more data than that to reach a conclusion. So a real success for objective measurement by mathematics. Congratulations.

( There is an explanation at http://www.gokgs.com/help/rmath.html , but your interactive experiment is perhaps more intuitively illuminating. )

I dont think anyone does

_________________
Stefany, web programmer

It sure has its mysteries.
The other day I played a Free(!) game against a [-]-ranked(!) opponent, and right after the game finished, my rank changed!
Granted, I was right on the edge between two ranks but still, this was unexpected.

I think it's pretty safe to say that this had nothing to do with the game you were playing, and everything to do with minute changes in your ranking based on the shifting ranks of your previous rated opponents, that just happened to line up with a rank barrier.

Yep. Ranks can (and have) changed in the middle of an ongoing game... multiple times. (This has happened to me, and I've seen it happen to others as well.)

ez4u I was just reading the description if the rating system on the KGS website, and if I understand it correctly, the system would assign you a rating somewhere in the middle.

There is also a certainty value that's computed. For six games, it should always be relatively low, but I'm not sure enough to say if those results would make it especially low.

_________________
Occupy Babel!

Since people were describing the system determining upper bounds based on losses and lower bounds based on win, I was just wondering what would happen if the we had a perverse situation where the lower bounds were above the upper bounds.

_________________
Dave Sigaty
"Short-lived are both the praiser and the praised, and rememberer and the remembered..."
- Marcus Aurelius; Meditations, VIII 21