Skip to content
Little GolemMake your next move.

Little Golem Monster Rating

Public forum discussion.

General

#51 · quake

10 Mar 2008

(ah, good Dvonn players.. now I'm biting;) )

When you start playing Go, you get a kyu of 20.9. I think this should be set to 1500. (and the increasing steps also translated to 16 points for a win against an even matched opponent, to a max of 32; but maybe a counter argument could be that a go19 game is much bigger than other games, so it would be reasonable that it earns more points)

Although I'm not against three Go rating (they are more or less different disciplines), having an average or max is not a bad idea to capture the problem, that some players have high initial go ratings just because they have set them to that value. There are a lot of players who are playing mainly just one go type, but also have high ratings of the other two go types. Sometimes only with just a few games and nevertheless a translated rating of more than 2000 points.

When most of the played games on this site where go games, I think there was no problem, but there are e.g. also a same amount of connect6 games.

#52 · Marius Halsor

10 Mar 2008

Quake: If it really was true that every player started with 20.9 kyu, it would definitely and undoubtably be a correct translation to set 20.9 kyu = 1500. However, as I pointed out in my far too long post above, this is not the case. People start with a self-defined rating between 20.9 and 1 kyu, which means that 1500 is somewhere in this interval. Those who start with 20.9, are mainly people who play no or very few GO games, which means that most GO games are played between players with a much higher start rating.

Ray: Yes, it's obvious that GO is weighted too heavily. And, as I pointed out in my above post, it DOES matter much if 20 kyu or 10 kyu is translated to 1500. Let's just compare four different systems, and look at players with 20k, 10k, and 1 dan ratings in all three GO games, and see how many points they get in total for their GO games in each system (the 20 kyu player gets 0 in each system, so let's just look at the other two):

A) Count all three GO games, 20kyu=1500

10kyu: 1500

1 dan: 3000

B) Count all three GO games, 10kyu=1500

10kyu: 0

1 dan: 1500

C) Average all games, 20kyu=1500

10kyu: 500

1 dan: 1000

D) Average all games, 10kyu=1500

10kyu: 0

1 dan: 500

I'd personally prefer system D, but can see why someone would like B as well. As you can see, when comparing systems B and C, 10kyu players gets more points in C than B, whereas the opposite is true for 1 dan players. Since there are more 10 kyu players than 1 dan players, I think one might assume that the translation is even more important than whether or not one counts all GO games, although, as I said, I would prefer system D.

#53 · quake

10 Mar 2008

Is it still the case for a new player to start with a self-defined rating or was it only something that was possible in the past? I can't remember that I could set my initial rating by myself. Maybe I've just forgotten that I had an opportunity then.

Why is it not possible to set 20.9 kyu to 1500? The possibility to self define your go rating on a higher rating (enhance experience from outside this site) can then be seen as for example starting with a dvonn rating of 1800.

Strange that it is only possible for go and chess. There are a lot of experienced new players of other games on this site who have to make their way to the top just by playing a lot of games.

#54 · Dvd Avins

10 Mar 2008

Only Go and Chess have well-established ratings systems in general use.

The biggest problem is triple-counting Go. That makes what Marius is correctly pointing out triply bad, but there would be a problem even if Marius's objection is met. IMO, your Go rating should be the best of your three board-size Go ratings and your Hex rating should be the better of your two board-size Hex ratings.

People who know more than I should say whether Connect6 and LG's version of Go Moku as similar enough to be combined into one rating, or are more like Hex and TwixtPP, which are properly (IMO) counted separately.

#55 · Tasmanian Devil

10 Mar 2008

Anyone with a poor go rating can stop playing with their current nick, register a new nick as 1 kyu and then embark on catching up with their old rating in the other games. This will make the comparison of go ratings meaningless and pollute the ratings in the other games. How to deal with this problem?

#57 · tasuki

10 Mar 2008

Just relax, people... no need to fight against go so much :)

(after all, go is (obviously) the best of all the games)

So I changed it to 10kyu = 1500, but I think that's discriminating against go a bit.

The fact that my twixt rating is very close to my go19 rating and better than my go13 and go9 rating is just laughable. A long time ago, I invested substantial amount of time to studying go, while I never really studied twixt. I could talk about go strategy for dozens of hours while I could talk about twixt strategy for about 10 minutes. So I think my go rating should be much much higher than my twixt rating. Well, I consider this 10kyu = 1500 to be compensation for the fact that there are three different flavours of go on LG :)

As for cheating... there are easier and way more effective ways of cheating than "register a new nick as 1 kyu and then embark on catching up with their old rating in the other games" :]

I already explained why I'm not going to only take maximum go rating into account. And the more I think about it, the more I dislike averages as well.

PS: If go players feel they are discriminated against, please say so (I can always change it back if there are enough supporters :-)

#58 · movieloverxxl

10 Mar 2008

wow, what difference! I'm up from place 249 to 68!

#59 · Ruben Berenguel

10 Mar 2008

I feel discriminated :) I have also played a lot (lot, lot) of go (but here I only play 9x9) and 10k -> 1500 is laughable. I would put it around 15k, at least. Anyway, I am not a "monster" as I only play a few games heh.

#60 · Tim Shih

10 Mar 2008

In my opinion, a meaningful Monster Rating (MR) will be:

MR = (w1*g1 + w2*g2 + w3*g3 + ... + wn*gn) / n,

where g(i) is our currently-calculated rating for a certain game type, and w(i) is a weighting factor.

Determination of w(i) should be related to the gaming time consumed by a typical game. For example, perhaps one month is consumed by a standard Hex19 game, whereas 3 days are consumed by a standard Four-In-A-Row game. If we use different weighting factors for both game types, MR will be, indeed, more meaningful, I think.

#61 · Marius Halsor

11 Mar 2008

Ruben: I have never played GO outside of Little Golem, and consider myself a very, very poor GO player. My current rating in GO9 (which is the only one I play) is 13.8 kyu.

Tim: Why would a game that takes long time be more important than one that takes a short time? I might understand it if we were comparing the number of games played, as it is obviously easier to play many short games than many long games, but we are not. We are comparing ratings/skills in the different games. To me, all games should count the same. The only sensible weigths, as I see it, is to either count all games equally, or to weight GO, Hex and Reversi with 1/3, 1/2 and 1/2, as they have several board sizes, but otherwise the same rules (which is what we do in the 2nd Monster).

Why should Go or Hex be more important than Reversi or Gomoku? It just becomes a matter of taste. Why should someones favourite game be more important than another persons favourite game? It simply makes no sense to me. And TIME? what if we were to include tic-tac-toe, but did like in Einstein - used several games for a match? Let's say 100 games in a match? Then it would probably be the longest game of all. Should it have the highest weight? Or you could include Hex 200x200. Weighting by time, it would probably outweigh all other games combined. Why would we want to do that?

Tasuki: Thanks for the change! Now the rating makes some sense to me, and would have been a good basis for seeding in the 1st Monster. I would, just like Ray, have preferred weighting Go, Hex and Reversi too, but at least now it's much better than it was. Comparing ratings in different games like this makes really no sense (unless used as a seeding for a Monster tournament or similar), but it sure is FUN with a "total" rating. If it were an "official" rating, I'd definitely lobby more to get the weights for GO, Hex and Reversi, but since it's not, I should probably stop babbling soon...

#62 · Tim Shih

11 Mar 2008

OK, all human beings born on earth are equal. If we do abide by this ideal, then we must be also fair by evaluating their achievements starting from the very moment when the baby lands on the delivery table.

But we do not seem doing that now. Hmmm, John has too many cousins. He gets too much help from his cousins to gain his today's achievement. So, let us count John's achievement starting from his senior year in high school only. What about his rough time during his middle school days when he had to get up early in the morning to delivery newspapers? Nah, we do not count that. :)

Before we all debate further, we must first draft up a clear objective of the Monstership (if it was already drafted up, I apologize). Is it to promote/reward/encourage players' devotion to:

(1) strictly diversities of game types? (If so, then a grade A in a clay-pot

making short course should weigh the same as a grade A in a calculus

course)

(2) playing skill?

(3) LG gaming time?

(4) cultivating the logical thinking abilities in general, thus our life is

further enriched? (If so, weighing Hex19 and four-in-a-row the same would

be inappropriate)

(5) ??

#63 · Ruben Berenguel

11 Mar 2008

Marius, I have been playing go for about 6 years. First KGS, real life and then here, as lack of time prevented me from the other two. I usually play very quickly, as I don't want to be more than 10 minutes or so playing. As I do with almost all other games here. In real life I was 14 kyu last time I entered a tournament, if I play again I'll do it as 11-12. Again, 1500 for 10k I think is underrated (at least).

#64 · klaashaas

11 Mar 2008

According to the ratings, my go is worse than my amazons, dots and streetsoccer. So, I guess I'm hopeless and should stop investing time in go. In stead, I should consider making a career in Twixt, since I've never devoted any time to it, and I'm almost a better twixt player than go player.

(It's sooo obvious that 1500<->10kyu is wrong. I used to be a reversi player. I know what it takes to reach 1900 points at reversi. If I compare my go skill with my reversi skill, there is no way that I'm 200pts weaker (24% winning perc.) )

#65 · Tim Shih

11 Mar 2008

To Klaashaas:

If you consider becoming a professional Twixt player, I will consider sponsoring you. :) How much, 150K Euros a year?

I do understand that some people may be red-eyed about the existence of 3 variants of Go. The following opinion is purely my own:

Go9x9 is basically the end game of Go19x19. So, if Richard takes Go9x9 off the site, I will not mind at all. On the other hand, some novice players may find it beneficial to learn games on larger boards later.

Go13x13 is quite different from Go19x19. I personally will NOT play Go13x13 games at all because playing them confuses my global strategies for Go19x19. For this reason, if Richard takes Go13x13 off the site, I do not mind at all.

Whenever a friend asks me for an advice as how to improve Go playing skill, I like to answer, "Do not play Go13x13."

For learning diversities of games, Go9x9 and Go13x13 are valuable in their own right, however.

#66 · Greck

11 Mar 2008

Tim, I want to become s professional Hex player... :D

#67 · Tim Shih

11 Mar 2008

Gregorio,

Your wish is granted, subject to the following condition:

Defeat all top LG Hex players with two handicap stones given to them by the end of 2008. :) I am sure you can achieve this goal if you are so determined.

#68 · Ruben Berenguel

11 Mar 2008

Go on Gregorio, go on! You can start giving me two handicap stones :)

Tim, don't suggest Richard to take out Go9x9! I like to play it here, as a 19x19 game takes quite a long time and I am a very poor tactician. Sure a lot of people here play 9x9 also for fun (it's more quickly enjoyable than 19x19, although with less possibilities)

#69 · Robin

11 Mar 2008

@tasuki

Could you add country statistics as well? Doesn't seem too difficult. If you feel like it and if you have the time to do it, that would be really cool!! :D

#70 · Greck

11 Mar 2008

I take the challenge :)

#71 · Tim Shih

12 Mar 2008

Ruben, I did not suggest anything. :) LG can have 100 variants of Go, and I still will not complain.

Gregorio, you like to take the challenge? very well. Read the following story first. :)

Once upon a time, a Hex student went to visit a Hex saint.

"Sir, I want to become a Hex pro. Could you take me?" the pupil asked sincerely.

"OK, I'll take you provided that you can only utter two words during a period of 10 years. And do so near the end of the period. Are you able to do that?"

The pupil nodded his head firmly, and so the saint accepted him.

Time flies like an arrow. After 10 years, the teacher summoned the pupil over,

"You may speak up now."

"Food cold."

"Very well. You have improved Hex skill. Go back to study some more."

Ten more years elapsed. The saint summoned the pupil over, again,

"You may speak up now."

"Bed hard."

"Very well. You have further improved. Now go back to study some more. Study those games played by Nie_wiesz and Bush, for example."

Ten more years quickly elapsed. Seeing the pupil in front of him, the saint asked,

"So, what do you have to say this time?"

"I quit!"

#73 · Greck

12 Mar 2008

Tim, very discorageous... it will not mine my determination :P

#74 · Greck

12 Mar 2008

and the pupil had 30 years!!

#75 · Ray Garrison

13 Mar 2008

I think that Go has been oversimplified at 10kyu=1500. For example, I started at 19kyu and have won about 59% of my games since then, but that only brings me up to around 12kyu and a rating under 1400. I am clearly better than 1400 strength in Go!! I am currently ranked at #249 out of 685 Go players, easily in the top half, which also suggests I should be solidly over 1500. Also, look at #2, Tim Shih. At an outstanding 6.7dan he only equates to 2240??? He should be more like 2500! I might suggest something in between. This is the conversion table that I created and used when I seeded the players into the monster2 tournament (it may not be ideal either, but I think it is more accurate):

3 dan or stronger 2400

2 dan 2350

1 dan 2300

1 kyu 2250

2 kyu 2200

3 kyu 2150

4 kyu 2100

5 kyu 2050

6 kyu 2000

7 kyu 1950

8 kyu 1900

9 kyu 1850

10 kyu 1800

11 kyu 1750

12 kyu 1700

13 kyu 1650

14 kyu 1600

15 kyu 1550

16 kyu 1500

17 kyu or weaker, 1450

#76 · Tim Shih

13 Mar 2008

Ray, are you suggesting that this conversion table be adopted for all the three variants of Go? or just Go19? Thanks for your dedication of creating this table.

And it is, indeed, impressive that you advanced to 12 kyu in such a short period of time. If you show us a photo of yours, I expect to see two heads sitting on your shoulders. :)

#77 · ypercube

13 Mar 2008

two heads? only? He is the Monster!

#79 · quake

13 Mar 2008

Sure, it's obvious that the current go rating is oversimplifying go.

I thought it was just a compensation to the domination of go in the monster rating. Ok, the contribution of go should be substantial (maybe more than other games, but that is reached by having three go games), but I think is was too great in the initial monster table of tasuki. All we want is a nice representation of the total achievements reached at littlegolem over all games.

The problem is not having three go games. I think the case that many play mainly one go game, but have three high go ratings (two of them just set to the level achieved at the first) was overdominating go in the first table. Although the first go ratings were also too high in my opinion. Ratings going up to 2700. I think such a rating is nearly impossible to reach in other games. (for example chess or hex, also complex games)

#80 · klaashaas

13 Mar 2008

... 2700. I think such a rating is nearly impossible to reach in other games. (for example chess or hex, also complex games)

This only means that those games are less complex than Go. Seems a plausible explanation for me.

I don't hear Einstein players complain that the top twixt ratings are too high, because they can't reach such ratings in their game...

#81 · Greck

13 Mar 2008

I am not sure about the "complexity argument". But i'm pretty sure there is a long tradition on teaching/learning go, and most of people who like go has taken lessons, made problems and/or read books. That makes that go ratings are far more expanded than those in hex, for example.

Furthermore, my experience says that Go is less prone to "surprises", and that's why we don't use swiss system in go, but McMahon. Maybe that's another reason why go ratings expand with time.

Anyway, those paragraphs were not intended as an argument nor counter-argument in this discussion :)

yours,

Gregorio

#82 · Marius Halsor

13 Mar 2008

OK, I believe most of us agree that comparing ratings in different games is really difficult, and in many ways makes no sense. However, since we decide to do this, I believe that most of us agree that since everyone starts with 1500 rating in each game (except GO), ratings should be compared directly (1700 ratings in Einstein compare to 1700 ratings in Hex).

But in GO we don't start with 1500 rating. If everyone had started with the same rating, say, 20 kyu, then I suppose we could have agreed to compare 20 kyu to 1500. However, since we DON'T all start with the same rating, that is not possible. An easy way to check out what the translation should be, is to look at the average ratings of all GO players, and compare them to the average ratings of the other games (which are close to 1500). The best way, however, would be to cronologically sort all GO games ever played here, and give all players a start rating of 1500, then go through all the games and find out what the current rating of each player should be. That's a bit of work, but since this discussion seems to attract a lot of interrest, I think I'll give it a try - after Easter. Just for fun, and to see what translation would have been "correct". I have some predictions, however:

- Players who started with a 20 kyu rating will get a higher rating than they have now

- Players who started with a 1 kyu rating will get a lower rating than they have now

- Players who have played lots of games will have a smaller difference between their "old" and "new" rating than people who have played few games

- The 1500 rating will be somewhere between 15 kyu and 8 kyu

The three first might be possible to prove. The last one is just a guess.

And Klaashaas, if anyone ever gets an Einstein rating above 2500, I will definitely believe they are cheating. I think that even with perfect play in every game, one could not reach such a rating, due to the stochastic nature of the game and the small pool of players.

#83 · Abigail

13 Mar 2008

Tim Shih, such a long joke, and then you still miss the punch line. "I quit" by the pupil is followed by the master saying "I'm not surprised. You complained for the last 30 years!"

#84 · Tim Shih

13 Mar 2008

Ahhhhh, Abigail, I did miss the punchline!! Thanks. So, Marius The Expert, how do we translate one's rating in telling jokes when the joke teller has even missed the punch line herself/himself? 1300? :)

#86 · FatPhil

13 Mar 2008

Tweak - "... You did nothing but complain for the last 30 years" ?

#87 · tasuki

13 Mar 2008

It seems that the majority agrees that 10kyu = 1500 is not putting enough of value on go. So I decided to put it on 14kyu = 1500 (and it should stay this way at least a little git closer, I actually think we might be (slowly) approaching a consensus :). This is still intentionally a bit lower than what I think is appropriate for go, but that's to compensate for the fact that there are three flavours of go here.

Ray: I see you have a gap of 50 points between 1kyu and 1dan... as far as I can judge from some LG graphs, the difference is in fact 5 points (it goes like: 1.3k, 1.2k, 1.1k, 1.0k, 1.0d, 1.1d, ...).

Robin: country statistics, mmm... do you mean like a flag next to a player's name or some kind of searching by country or something else?

I'd like to create customizable ratings - ie. you choose which games to include and what weights to put on them, but now I really have no time for that at all (I'm going skiing for a week, then I'll have catch up at work and then... who knows ;-)

#88 · tasuki

13 Mar 2008

I meant to write "a little bit closer", of course...

(why do I never notice typos when rereading the post and then see them 2 seconds after submitting?)

#89 · Dvd Avins

14 Mar 2008

THe compensation of having three flavors does me no good, as I'm a little below "1500" in all 3. Nonetheless, I think my 15k Go rating reflects more understanding and strength than my 1600-1700 Hex13 and Amazon ratings do, though not as much as my 1700s Chess rating does.

#90 · Ray Garrison

14 Mar 2008

To clarify, the go conversion chart I gave above I had worked out last fall in order to seed the players into Monster2 (I am the tournament director). For Monster2, Go games are weighted and count only 1/3 and Hex and Reversi games count only 1/2. Therefore, I averaged players ratings beforing converting to a rating. I never got any feedback on my conversion table at the time, so I just went ahead and used that. However, for the Monster2 finals, I may tweak the chart a bit. As for the gap between 1kyu and 1dan, I was very unsure about how that gap actually worked. I thought it might aproach 0 from both ends, I never even considered that it jumped from 1kyu to 1dan (skipping .9kyu, .8kyu....0...., .8dan, .9dan), I will adjust my table accordingly for the next round of games (thanks tasuki!)

#91 · Ray Garrison

14 Mar 2008

P.S., tasuki, I think 14kyu=1500 is a good value

#92 · MarleysGhost

14 Mar 2008

Another approach to Go ratings on LG would be to simulate what would have happened had Go used the same start-at-1500 chess-style ratings as the other games. While the results of every game ever played on LG are apparently available online, to implement this suggestion, you'd also want to know the order in which the games ended, and AFAIK that can only be approximated from the online records.

In any event, if you had those simulated ratings, you could make a scatter plot of them vs. the existing Kyu/Dan ratings and see if there isn't a natural linear relationship.

#93 · tasuki

14 Mar 2008

Ray: thanks for sharing the opinion that 14kyu=1500 is good :)

MarleysGhos: That could be useful, but it would be hundred times more difficult to implement than the current (simple) system. (You know, sometimes I like to pretend I have a life too ;)

#94 · MarleysGhost

14 Mar 2008

Here's another, slightly easier, approach: For each game, find the mean and standard deviation of the ratings of players meeting a criterion (all who have ever played, all who have played recently, all who have played more than N times, whatever). If there isn't too much scatter in those values among the different games, then it makes sense to assign rating points to the Kyu/Dan values such that the resulting mean and standard deviation are close to what they are for the other games. Unlike the time-history of all game outcomes, this could be easily automated. In fact, hasn't someone already fetched the ratings of all players in all games?

Of course, if the different games have wildly different means or deviations, selecting which games to match the Go ratings to would still be an open question.

#95 · Robin

14 Mar 2008

Robin: country statistics, mmm... do you mean like a flag next to a player's name or some kind of searching by country or something else?

I meant something like this.

#96 · Marius Halsor

25 Mar 2008

OK, for GO9, I have now gone through every game ever played, and gave all players a start rating of 1500. I have then calculated the current rating for all players. Looking at the top of the "monster rating" table, here are some examples (monster rating first, then my calculated rating). I start with players who started on LG with a high rating:

Bernhard Herwig: 2075 - 2035

Tasuki: 2210 - 2108

Xavier Gubas: 2260 - 2126

Bela Nagy: 2145 - 1968

Nosferatu: 1735 - 1646

Then players who started with a low rating:

Ray Garrison: 1575 - 1665

Klaashaas: 1880 - 1853

PAKO: 1290 - 1577

Marius Halsor: 1510 - 1659

As I predicted, players with a high start rating got a lower rating this way, and players with a low start rating got a higher rating this way (with an exception for Klaashaas - possibly because he has played very many games, or possibly because his games have mainly been against players who had a high start rating).

I do NOT suggest you use this method, as it is not intuitive for most players, and is also a bit time consuming. However, I think one can see from this experiment that the high ranked GO players still get too high score (while low ranked players get too low score). Some might argue that this means we now have the "right" translation, as the difference for high and low ranked players are approximately the same. However, as only ratings above 1500 are counted, it is the rating of high ranked players that is important. I therefore think that the translation still is a bit off (maybe 12kyu=1500 is better?).

Anyway, it's not off by more than a 100 points or so. The fact that there are three GO games (and two Reversi/Hex games) is much more important now, than what I see as an error in the rating translation.

Oh, and I would have done this for GO19, but was not sure how rating changes are calculated for MC (handicapped games). But the principle is just the same, and the rating changes will be similar.

#97 · movieloverxxl

25 Mar 2008

80 lone 1890 2130 2130 2130

If you have a look at this player: some while ago he set hisself as 1.4kyu in all 3 go games. He only played friendly games and never 'proved' his rating. IS there no way to exclude such players as he never participated in any tournament ever (but is 80th in the Monster Rating...)

#98 · Marius Halsor

25 Mar 2008

There was a slight error in the previous calculation (ongoing games got a weird value), but numbers are similar. I'll post them for completeness:

Bernhard Herwig: 2075 - 2148

Tasuki: 2210 - 2108

Xavier Gubas: 2260 - 2234

Bela Nagy: 2145 - 1968

Nosferatu: 1735 - 1645

-----

Ray Garrison: 1575 - 1725

Klaashaas: 1880 - 1915

PAKO: 1290 - 1579

Marius Halsor: 1510 - 1697

With this corrected, the fit with 14kyu=1500 is better than I thought before noticing this error. I have not looked at more players than these nine so far, but looking at this population, I think the current translation factor of 14kyu=1500 is OK.

#99 · bloke

25 Mar 2008

i still do not quite understand why results and rankings the game of go are so much overrepresented. why not calculate with go9, go13 and go19 as one game with one average ranking for each player? why not use the same for hex13 and hex19 as well as for reversi8 and reversi10?

if someone is an excellent player in amazons, gomoku, connect6 or streetsoccer and above verge in all others then that person is lower ranked than some go players who have not even tested themselves in any other games.

why is this issue not discussed seriously?

otherwise congrats tasuki, excellent site

#100 · tasuki

25 Mar 2008

As has been said, go19, go13 and go9 are different games. So are hex13 and hex19 (and I have every reason to suppose that so is reversi10 and reversi-whatever).

You could also argue that draughts and mancala games are horribly underestimated in my ratings ;-)

As for people setting their go rating and not playing... well... that is a problem. It's a much bigger problem than the difference between ratings counted from scratch and ratings counted from some artificial starting value. It's also a problem that is not easy to solve. I believe that any "solution" of this problem will make many people angry. Taking into account the number of rated games might be a solution, but I'd have to set it to some kind of arbitrary value anyway.

And as for go players being overrepresented at the top - Nao is the best "go-only" player, currently ranked 36. I think that's rather reasonable.

And I promise that when I get the time, I'll implement customizable ratings - you set the cutoff value and the weight for each game, so everyone will create their own rating and will be satisfied. :-)

Last page