Why such divergent scoring results with different GUIs?

Alberich · Post by **Alberich** » Sat Jul 07, 2012 6:09 am

I played a game this morning against GNUGo using CompoGo. After the game was over, CompoGo scored it as a win for me by 11 points for White. But when I imported this game into Smartgo Kifu using Dropbox...Smartgo Kifu scored it as 236 points?! for Black. They can't both be right. Is this issue with scoring a big problem for go software?

ez4u · Post by **ez4u** » Sat Jul 07, 2012 6:50 am

Alberich wrote:I played a game this morning against GNUGo using CompoGo. After the game was over, CompoGo scored it as a win for me by 11 points for White. But when I imported this game into Smartgo Kifu using Dropbox...Smartgo Kifu scored it as 236 points?! for Black. They can't both be right. Is this issue with scoring a big problem for go software?

Which one did you think was correct? :blackeye:

Alberich · Post by **Alberich** » Sat Jul 07, 2012 8:04 am

I don't know...because right now Igowin HD doesn't support iTunes file sharing nor email importing since that app I trust more than Smartgo Kifu. So I don't know why different GUIs are giving such different results. I'm surprised there isn't a uniformed method of scoring implemented across the different programs.

SpongeBob · Post by **SpongeBob** » Sat Jul 07, 2012 12:03 pm

At the end of the game, it has to be decided which groups are alive and which are dead. It is not so easy for the computer to determine this correctly. When you play on KGS for example, you and your opponent have to agree upon and mark the dead groups manually.

If the status (dead or alive) of all groups is decided and all moves that yield points have been played by the players (which is the case, normally), the result will be the same for all implementations.

Alberich · Post by **Alberich** » Sat Jul 07, 2012 4:47 pm

That's incredible. Can you imagine fierce enemies on the board sitting down to calmly argue which side's stones are dead? Western chess doesn't have this problem because not only is the board smaller but the goal there is clearer...capture or kill the opposing king. No such goal exists in Go. The end of a game between two players looks like a work of art (it really does look like art) and each player determines which stones are dead. If that's the case here then there's no engine or GUI out there that can reliably determine who's the winner in these games. KGS does a good job of this but unfortunately it doesn't have a built in engine that does this when you play against the computer when not connected to the KGS server. It's only available online. In second place I'd have to award this to Compogo. I think this GUI does a very good job of figuring out the proper counting of the stones.

Does anybody know if Compogo is still being actively developed?

hyperpape · Post by **hyperpape** » Sat Jul 07, 2012 5:54 pm

It's worth noting that there are rulesets where life and death are always settled and can be mechanically checked. I believe computer vs. computer matches use a method like this. At the end of the game where humans would agree on life and death, the computers fill everything in until stones are captured.

In fact, in roughly 90% of human games, there's no question of agreement. Both players can see without a problem that the game is over and which territories are which. In another 9%, one player has forgotten a single dame (a neutral point that is of no particular significance). In perhaps 1%, someone really has forgotten something and loses points or there is a disagreement. Just today, I lost ten points because I'd overlooked a move inside my territory when I passed. But even there, there was no argument. By opponent played it, and I lost several stones).

It is only more human friendly rules that cause problems for computers by permitting us to omit moves at the end of the game. Btw: it's not just computers that can be confused. In a game between two 10 kyus, a corner position may be played out until the bitter end. Between two 9 dans, they may simply know better than to play any more. (Similarly, it's much less common, but two weak chess players will not know that a particular endgame position is a draw).

jts · Post by **jts** » Sat Jul 07, 2012 7:59 pm

I think you're conflating two different problems, alberich.

1: of smartgo kifu and compogo, one (or both) of them made a mistake scoring a position.

2: two beginners will sometimes make mistakes in the endgame.

The first problem is sort of banal... Next time you have this problem save the sgf and we will tell you which program is counting correctly, and why. Or learn to count yourself, so you won't have to rely on the computer to tell w+11 from b+250

The second problem is more equivalent to two chess novices each missing several checkmate sequences before the game finally trundles to a close. If you find that sort of sloppiness aesthetically off-putting, play more go and you'll encounter less of it.

HermanHiddema · Post by **HermanHiddema** » Sun Jul 08, 2012 11:07 am

Alberich wrote:That's incredible. Can you imagine fierce enemies on the board sitting down to calmly argue which side's stones are dead? Western chess doesn't have this problem because not only is the board smaller but the goal there is clearer...capture or kill the opposing king.

Can you imagine fierce enemies on the board sitting down to calmly agree that K+R vs K is a won game?

As they say: Chess is a battle, go is the whole war. Go consists of many individual battles, and at some point, in each of them, there is no point in playing on. Skilled players know this, and will not play on. Just as any reasonable chess player knows that K+R vs. K is won, any reasonable go player knows when stones are dead. In either case, such players will not needlessly play on, they will concede the position.

xed_over · Post by **xed_over** » Mon Jul 09, 2012 1:07 am

Alberich wrote:That's incredible. Can you imagine fierce enemies on the board sitting down to calmly argue which side's stones are dead?
...
No such goal exists in Go.

play using Chinese rules, then the one with the most stones on the board wins.

RobertJasiek · Post by **RobertJasiek** » Mon Jul 09, 2012 11:15 pm

Alberich wrote:then there's no engine or GUI out there that can reliably determine who's the winner in these games.

It is no problem for area scoring.

KGS does a good job of this

For KGS-Japanese Rules, it makes simple mistakes, i.e. it does make a terrible job.

RobertJasiek · Post by **RobertJasiek** » Mon Jul 09, 2012 11:20 pm

xed_over wrote:play using Chinese rules, then the one with the most stones on the board wins.

Chinese rules are not the solution because they contain ambiguity. Use simple area scoring rules instead.

Chinese rules do not let the one with the most stones on the board (stone scoring) win, but use area scoring.

HermanHiddema · Post by **HermanHiddema** » Tue Jul 10, 2012 2:54 am

RobertJasiek wrote:
Alberich wrote:then there's no engine or GUI out there that can reliably determine who's the winner in these games.

It is no problem for area scoring.

RobertJasiek wrote:Chinese rules are not the solution because they contain ambiguity.

Chinese rules...use area scoring.

There seems to be a slight contradiction here, no?

More to the point, it is a problem for any rule set that has a agreement phase.

RobertJasiek · Post by **RobertJasiek** » Tue Jul 10, 2012 3:49 am

HermanHiddema wrote:There seems to be a slight contradiction here

Of course not. The Chinese Rules' major problem is the ko rules.

Bill Spight · Post by **Bill Spight** » Tue Jul 10, 2012 9:46 am

This is a problem for beginners. It would be a problem for kyu players if they did not agree through ignorance. (A small percentage of kyu games are incorrectly scored.) It is not a problem for pros, which is one reason that go rules are sloppy. We cannot now expect GUIs to score at pro level.

Life In 19x19