Autoscore / Score Estimate problem

In this game auto-score have scored C2 group as dead. This problem is known, but there is another one. If I take final position and select “Estimate score”, I can’t make it mark all stones as living either:

Correct score is 28.5, as far as I understand but I can’t seem to make the tool display it.

Let me help you clarify some things :slight_smile:

First, “estimate score” is unrelated to counting. It is not designed to give exact results. Please disregard this feature in discussions on counting [a final position].

Next, the correct score is B+30.5.

  • Black has 34 points of territory.
  • White has three prisoners and 0.5 komi.
  • Eyes in seki are not points in Japanese rules.

With that out of the way, I don’t understand what exactly you are reporting. I read three meanings:

  1. Are you unsatisfied that live and dead stones are not automatically marked correctly, and that the eyes of the seki are not automatically marked as neutral points?
    Please be aware that it is the responsibility of the players to correctly mark the status of the groups at the end of the game. Normally, you click on a group to toggle it between live and dead. Click on a space to toggle it between territory and neutral. Do this until all stones in seki are marked live and all spaces in the seki are marked as neutral points.

  2. Are you bothered by the fact that (I believe) you cannot negotiate with the bot on the final status of the stones? The bot gets to mark the stones, even if it violates the rules.
    This is because on OGS, we do not trust all users to score their games in a fair manner against bots. Also, it is impossible to accurately determine the status in all cases automatically.
    You could try to report the board position to the developers of GNU Go, but that project is based on outdated algorithms and long abandoned.
    You can negotiate group status with some bots if you run them on your local machine. This is definitely the case with GNU Go. But… seeing as you already finished 15 games against GNU Go, in the interests of your own Go strength and style, I recommend that you play against humans instead :wink:

  3. Did you have a case where you marked everything correctly, but the final score came out wrong? This could be a genuine bug. To be sure, I just did an experiment with a friend where we built this kind of board position and scored it. We found no issues.

1 Like

First, “estimate score” is unrelated to counting. It is not designed to give exact results.

So, «Estimate score» function uses different algorithm to count final position then one which used in the end of game? What is the reason to have two instances of code which does same thing?

Either way, I believe that inability to mark stones alive or dead is deficiency of this tool and should be fixed. I don’t see any problems in having «Estimate score» in final position calculate score exactly same way as actual scoring function which is activated after two passes.

Next, the correct score is B+30.5.

It is according to Japanesse ruleset, but games vs bots are scored under Chinese rules:

Black has 54 points of area: 51 for main group plus 3 for seki; white has 10 points for right group and 15 for top group plus 0.5, so it is 25.5 total. The difference is 28.5.

  1. Yes, I am unsatisfied by that too, but there is separate topic for this issue. I believe humans should have ability to select continue playing if they think stones are marked incorrectly.

  2. I thought that it is OGS auto-score function, which scores the final result (auto-scoring feature), not bot. Are you sure that it’s bot?

  3. Can you share this game?

I am bothered by the fact that I can’t mark dead/living scores in this particular position, while score estimate function does allow to mark stones alive or dead in principle. I believe this is a bug…

Estimate Score is a monte-carlo function that approximately computes the final score from the current board position at any move, even before the game is finished. My advice is that you should never use it.

Counting is a phase at the end of the game in which the result of the game is determined. It requires someone to judge the life and death status of the groups on the board. There is no feasible way to have the computer do it. That is why the players must mark the groups.
In case of disagreement, you are supposed to clarify the circumstances through negotiation with your opponent. You can even resume the game, if necessary.
You can probably not negotiate with a bot. You have to accept its judgement for technical reasons. I never play against bots, so I do not know first hand whether this is really absolute, but if you tried to negotiate and were not allowed, I would say it is because your opponent was a bot.

Point taken on your game being scored in Chinese :slight_smile: In that case, the score would be B+27.5. White can even fill in the fake eyes for a total of four points difference compared to Japanese (black also made the last move inside his own territory, so we arrive at three difference).

There is no “auto-score function” on OGS or on any other Go server. When you see some groups automatically marked dead in counting (in regular human-human games), that is merely a suggestion and a convenient shortcut that saves you from the menial task of clicking every single dead stone on the board. It covers most common life and death scenarios. Your seki is more complicated, so please pay attention and don’t “accept score” without thinking :slight_smile:

Here is the game where we re-constructed a seki just like yours and scored it under Japanese rules without any problems: https://online-go.com/game/11456315
It is private, but I gave access specifically to you.

I hope this helps to clear up some misunderstandings :slight_smile:

Estimate Score is a monte-carlo function that approximately computes the final score from the current board position at any move

Yes, but it also allows selecting dead stones.

My advice is that you should never use it.

What is function for if it should never be used?

Counting is a phase at the end of the game in which the result of the game is determined. It requires someone to judge the life and death status of the groups on the board.

The thing is that players sometimes pass even if game is not finished, such as dames are not filled or even if there is an open peek into opponent’s territory. Auto-counting at the end of game should also be able to compute scores for arbitrary board positions, even if it’s mostly used for fully enclosed positions.

When you see some groups automatically marked dead in counting (in regular human-human games), that is merely a suggestion and a convenient shortcut that saves you from the menial task of clicking every single dead stone on the board.

Isn’t it using same monte-carlo algo as Score Estimate does?

the score would be B+27.5.

I thought it was 28.5, how did you calculate it?

I summed up my opinion of the score estimator in another thread recently :slight_smile: , so I will simply link that: Score Estimate Bug?

And where do you get this idea that unfinished boards should be scored? What if an unsettled area makes the difference between a black win and a white win? It is impossible to decide and also highly unusual.

I don’t know how dead groups are pre-selected, actually. My unsubstantiated gut feeling is that it is not random. I guess it is an interesting question for the developers :slight_smile: , but also practically irrelevant.
Whichever method, it only needs to be “good enough” to cover the most common cases.

I got the score wrong twice. It is B+28.5, like you said all along :joy:
Then why is there such a weird difference to the territory score of B+30.5, when the two methods are supposed to be basically equal? This threw me off until I realized that this is a handicap game; black has one extra stone at the beginning of the game (in addition to the last move).

What do you suggest server should do if both players pass before territory borders are settled? Of course players should not do that normally, but I have seen them doing it.

I think score estimator and endgame score calculator should use same interface and functions. For example I use “Esitimate Score” when player resigns in yose, then I click through obvious yose sequence and select “Estimate score” to see what score would be if there was no resignation. And in some games players even resign in finished position.

I would suggest that they learn how to play go, and to recognize unsettled territory borders. :wink:

Hey there,

I had a similar Issue playing against GnuBot 5k.
I’m fine with GnuBot doing the scoring in the end, however I’m kind of confused that when I wanted to find out the correct result during analysis but I wasn’t able to mark the stones in seki as ‘not white territory’. I’m pretty new to OGS and its system for analyzing games so there is a chance that I simply got the controls wrong or didn’t use the right settings.

This is the game:
[https://online-go.com/game/12049074]