Poll: how would you score if you were the auto-score algorithm?

Here’s a few games. Both players did not resume the game in each case (e.g., they just left), and it’s now up to you, the auto-score algorithm, to decide who wins the game.

Feel free to comment below if you want to explain your answer, but keep in mind that you’re a bot: be consequent, and know that in practice you can’t explain your answer, or give nuance.

[edit: I’ve included sgfs, and uploaded them to ogs, see the links below the pictures.]

Game 1

link

  • Black wins
  • White wins
  • Game is annulled

0 voters

Game 2

link

  • Black wins
  • White wins
  • Game is annulled

0 voters

Game 3

link

  • Black wins
  • White wins
  • Game is annulled

0 voters

Game 4

link

  • Black wins
  • White wins
  • Game is annulled

0 voters

Game 5

link

  • Black wins
  • White wins
  • Game is annulled

0 voters

5 Likes

As for scoring areas with unclosed borders:

Where would you draw the line if you want to award points for an unclosed area? How many gaps would you allow and how big can those gaps be, before you award no points for an unclosed area?

2 Likes

Anything not alive yet is dead? Or anything not killed is alive? How would a bot think then there are two killable groups by W and B can only save one?

When both players pass, they should mark what’s dead. If they agree on that, it’s dead, regardless of what a spectator, stronger player or AI might think.

Only if they disagree, it might become complicated.

So in the poll, we’re missing which stones were marked dead by the players. If that is left to autoscore and the game is incomplete, the autoscore algorithm needs to be quite smart and almost be able to read the minds of the players.

4 Likes

When i open game 2 the diagram shows the upper right like a seki (which is obviously not )
Is that a bug ? I guess It’s related to the missing move P 16.
I don’t really understand the sgfs for other games, what are they showing, the scoring?

The sgfs show the automatic marking of the territory where all stones are considered alive. It’s not really relevant to the question.

As autoscore i should’nt play at the place of the players to finish the boundary, decide if there is a move to add or not and determine what is alive dead… So the only reasonable decision, if there is no way to have the players resume playing, is to annul all (or to refuse scoring at least )

1 Like

What about normal dame points, they generally aren’t filled by players. How do you find the difference between a hole in the territory and harmless unfilled dame?

And what about the last two games? Presumably the players didn’t think that the game was unsettled in either case.

What rules are we supposed to be using? I assumed it would be Japanese, but the links say Chinese.

Begin bot :robot:

I am bot. I cannot explain nuance. But I can state my algorithm:

  • All stones on the board are assumed alive unless players tell me that they agree they are dead.

  • Empty intersections connected to stones of one color only are territory of that color.

  • Empty intersections connected to stones of both colors are neutral.

  • Alive stones touching neutral points are in seki.

  • Under Chinese rules, territory surrounded by stones in seki count as points; under Japanese rules, they do not.

End bot :robot:

(Note that in the OP, all games are Chinese rules and 5.5 Komi)

Results

Game 1: B+18.5

Game 2: B+35.5

Game 3: W+41.5

Game 4: B+12.5

Game 5: W+6.5

Simple algorithm. No need to judge life and death. No need to judge open borders. Game is scored as it is. Never fails. If players choose to:

  1. Pass before borders are closed and/or before dame are filled; and

  2. Also leave without marking dead stones

then that is their free choice but don’t be surprised if the auto-score does not guess where you would have played! (How can it?)

Players must do at least point 2 above (or do point 1 plus capture all dead stones) to ensure auto-score is as they want it automatically.

5 Likes

This is why i recommend the marking of the boundaries by the players. Then there is an agreement. Without marking you give the power to the AI to play for you whatever happen.

For the dame it’s the same and i have the reverse impression that players tend to play them more and more. At least It’s no more a sign of disrespect and my guess is It’s because of the emergence of the Chinese rule and the bots.

I don’t think it matters, here, but if you think it does, feel free to say so.

2 Likes

I voted according to my Proposal for a Simple Ruleset.

This makes all the situations quite clear and it doesn’t matter what the players were thinking or agreed on. :slight_smile:

The difficulty is in actually counting and reading like KataGo. I’m not at that level, but I think I got all the tsumego right. :sweat_smile:

2 Likes

The sgf’s include a level IV analysis by KataGo, if you need it :slight_smile:

2 Likes

Might do, might not. Depends on the auto-score algorithm and the games themselves.

Under the simple algorithm I proposed above (Poll: how would you score if you were the auto-score algorithm? - #11 by dragon-devourer), where points in seki count in Chinese rules but not in Japanese rules (as per the actual rules), it is unclear whether the ruleset would make a difference without checking so here we go:

Game 1: Changes the winner (B+18.5 → W+2.5)

Game 2: Same winner, different score (B+35.5 → B+36.5)

Game 3: Same winner, different score (W+41.5 → W+15.5)

Game 4: No difference (B+12.5)

Game 5: Same winner, different score (W+6.5 → W+8.5)

So, game 1 shows ruleset does make a difference.

3 Likes

But, I’m very glad we don’t have your algorithm :rofl:

2 Likes

But why not? It’s simple to understand, simple to implement, and encourages proper play. People might argue that it scores incomplete games “incorrectly”, but if the game is incomplete then what do you expect!? Provided people, as a minimum, close borders, and agree on dead stones, dame and teire, then it works fine. Since all those things are part of the game, then I can’t see what the objection is.

1 Like

Using 2003 Japanese rules as written results in whole board seki in most cases.
Game 1: The only group not in Seki (not possessing dame) is the black top left group.
Game 2: 3 black groups are not in Seki, again white has no territory at all.
Game 3: Probably no territory at all => score=komi (+captures)

(For Tromp-Taylor rules => no stone removal phase. )

3 Likes

I’m too lazy to count my own games, leave alone theoretical exercises where the goal isn’t clear. :joy:

I feel like the question is missing some few statements. For instance: do you want a smart or a dumb bot?
Maybe that’s actually the question, as emerged in another thread.

Do you mean pass + pass + just leave?
Well, my opinion is that if they don’t care scoring, then they don’t deserve score: game annulled.

This actually isn’t “a bot”. It’s a simple rule in the UI: players have to close borders, mark dead stones and agree on the result. If they don’t, then the game is annulled.

2 Likes

Color go server has interesting visuals, with the idea that every stone emits influence:

Screenshot

I don’t know the details, but how about using an algorithm inspired by this? Something like the following:

For every intersection, calculate a weighed sum over all other non-empty intersections, with the weight being 1 / d ^ 2 , where d is the distance, and taking different sign depending on colour.

Once this weighed sum is calculated for all intersections, if the result is too close to zero it counts as neutral point, and otherwise as territory for the respective side.

Definetely not a rigorous scoring algorithm but could be a fun idea to play around with.

1 Like