Go Style Estimator

[Bill Spight]

I'm 7k, I guess. And I like to capture stones.

Oh, don't be so tentative. More machismo, if you please. Like this:

Hi. I'm Seven Kyu. {Pause to let that sink in.}

I capture stones.

[Charlie]

Interestingly, 37 recent games on IGS (going back to December, last year) put me at 2 kyu:
http://gostyle.j2m.cz/WAPP_RES/c5fbd3b8 ... e8fe8c460/
(I was actually 11 kyu to 9 kyu, IGS, for these games)

But 40 recent games on KGS put me at 5 kyu:
http://gostyle.j2m.cz/WAPP_RES/bea1a707 ... b8b33c5aa/

Both of the feedback pages instructed me to concentrate on sente and gote but the style statistics appear completely different.

[pasky]

Interestingly, 37 recent games on IGS (going back to December, last year) put me at 2 kyu:
http://gostyle.j2m.cz/WAPP_RES/c5fbd3b8 ... e8fe8c460/
(I was actually 11 kyu to 9 kyu, IGS, for these games)

But 40 recent games on KGS put me at 5 kyu:
http://gostyle.j2m.cz/WAPP_RES/bea1a707 ... b8b33c5aa/

Both of the feedback pages instructed me to concentrate on sente and gote but the style statistics appear completely different.

Hmm, to me it seems the statistics actually are quite similar except for "orthodoxity". The style statistics are influenced also by your opponents, of course; different competing styles may provoke you to play different moves, perhaps we could conclude that IGS players have more traditional style than KGS players. Or it's just a statistical error, the confidence intervals are unfortunately very wide still.

Thank you all for trying it out!

[Boidhre]

So I tried it: http://gostyle.j2m.cz/WAPP_RES/da8bfcab ... 6e489c5c7/

I've never played a whole lot of go online (and not a whole lot of even games) so it was never going to work right but a few things:

It says it took games from September 2012 to December 2012. It didn't, it took games from March 2012 onwards. This presents a problem in that it's mixing games where the server thinks I'm 22k or whatever right back when I started playing go, in with games where the server thinks I'm 8k. You can't really take games from when someone is just starting out in go and use them to give tips for play 9 months later. If the problems persist from the early games then that's an excellent thing to bring up (and if you do this you should highlight this!). I think there should be some check in the program to make sure you're not using games from a region exceeding 5 or 6 stones in strength. Also over a 14 stone strength difference (I don't actually believe I was 7/8k but whatever) any discussion of style over that series of games will be meaningless. I didn't know what thickness meant at 8k but I knew even less about it at 22k and so on.

Also the first game is wrong, I was 30k in that game, it was my first game of go and I thought 30k was appropriate. The problem could be on KGS's end though as my rank would have been ? for that game with 0 games played so no predicted rank. Is it possible this game being marked as 8k threw off your algorithm?


Nice tool though, good work.

[JoeM]

Boidhre: The wrong range of games is a clear bug, I'll fix this The rank of the player in each of the games listed is taken from the sgf file and is not taken in account in any way - it is listed just for convenience. The number of games from KGS is taken as the last 40 games currently, so if you have played less then 40 games on KGS altogether, it takes all of them. This is done (taking last 40 games) so that the algorithm has enough information, but in these extreme cases (improvement of 15 stones within the sample) this additional information might simply not be worth the error introduced by the difference in strength. However, I still consider the fact that the rank is estimated approximately as the average in the sample and the "Bad things you play" gives a long list a pretty good result.

Charlie: Actually, 9-11 kyu IGS (as listed in the sgf files), corresponds to 6-7kyu KGS. Moreover, the KGS range in the results is longer - February 2012 to March 2013, so that even the games where you were KGS 10kyu are included. So the results seem to overestimate your strength by 4 stones but consistently; similarly, the "Bad things you play" suggests in both cases that you should study sente/gote and focus on keeping the initative.

So maybe your overall gameplay is really stronger, but your main weakness is the sente/gote - once you fix this you could easily become 4stones stronger.

[Boidhre]

I've played 148 games on KGS. Not many but quite a bit more than 40. What kinds of games do you exclude from the sample? Free? Handicap? The latter makes sense to me, the former is awkward because you don't know which free games were serious and which were teaching games or whatever.

A long list of mistakes being reported is good. Not knowing if it's pointing out mistakes made at 20k that I'm no longer making, not so useful. I'll try and dig up a sample of games I recorded and played offline with a more consistent strength rating and see what happens.

[Boidhre]

Small sample, I fed it 60 odd games from August to December but it only chose 16 for analysis: http://gostyle.j2m.cz/WAPP_RES/f8957394 ... 67ce0b99c/ (I play a lot of handicap go offline since there's no one for me to play even games with)

Very different results. Thought you might find this interesting.

[JoeM]

Well, we only take into account games played on 19x19, without handicap stones (moreover, the sgf has to be valid and have only one game tree inside, but this is a technicality). The reason is obvious, smaller boardsizes and handicap go have "different" flow of the game and strategies - these cases would need to be treated separately. Of course, this would be really interesting to do and compare the results (e.g. style differences for handicap games, etc.), maybe in the future

[kwhyte]

I tried it with some of my games and got 3d +- 2.63, which is a bit high (I'm probably right on the border of 1k/1d). Out of curiosity I also tried it with a collection of 40 or so games of Go Seigen that I had in a collection, and it also ranked him 3d +- 2.63. It did correctly identify that Go Seigen was one of the pros with a similar style to him.

[JoeM]

Well, the algorithms are trained on KGS players from 20kyu to 6dan - and this is also the range of the strength scale. So the professional players (or players stronger than 6d) cannot even be classified correctly. On the other hand, the style should fit well.

[Pippen]

In my case it says: "The KGS equivalent of strength of the player: 3d ± 2.63". What does that mean? I am an 1k KGS for years now. Does that mean that I am a 3d in EGF or where?

[Boidhre]

In my case it says: "The KGS equivalent of strength of the player: 3d ± 2.63". What does that mean? I am an 1k KGS for years now. Does that mean that I am a 3d in EGF or where?

It means 3d KGS I think from the wording and what's been said on here. It's out by around 3 stones for me as well, I'm closer to 8k than the 5k it came up with.

[Pippen]

In my case it says: "The KGS equivalent of strength of the player: 3d ± 2.63". What does that mean? I am an 1k KGS for years now. Does that mean that I am a 3d in EGF or where?

It means 3d KGS I think from the wording and what's been said on here. It's out by around 3 stones for me as well, I'm closer to 8k than the 5k it came up with.

Well, but then the calculation is just wrong. I checked other players who have pretty stable ranks and they were all judged about 2-3 ranks stronger which is a contradiction to their actual ranks over the years and after thousands of games on KGS. It seems contradictory to say: You are a 3d on KGS while I am a 1k on KGS...or maybe it means: you could be a 3d if you played a little bit better?

[Boidhre]

In my case it says: "The KGS equivalent of strength of the player: 3d ± 2.63". What does that mean? I am an 1k KGS for years now. Does that mean that I am a 3d in EGF or where?

It means 3d KGS I think from the wording and what's been said on here. It's out by around 3 stones for me as well, I'm closer to 8k than the 5k it came up with.

Well, but then the calculation is just wrong. I checked other players who have pretty stable ranks and they were all judged about 2-3 ranks stronger which is a contradiction to their actual ranks over the years and after thousands of games on KGS. It seems contradictory to say: You are a 3d on KGS while I am a 1k on KGS...or maybe it means: you could be a 3d if you played a little bit better?

I put in a few people I know and they were all overestimated in rank if it's KGS rank that is estimated.

[Alexfrog]

Is the app still working?

Every time i try it recently I get Error 500 - Internal Server Error.