How does estimate score work? What is score estimation confidence?

OGS Development
1

I’d love to know more about the estimate score function and what score estimation confidence means.

3 Likes
2

It’s a very basic monte carlo estimator, the score confidence is the sum of fractional points awarded based on an average of all of the play throughs made during estimation. (Higher should be better).

2 Likes
3

This score estimation is one of the best (if not the best) I have seen on any go server.
There are just few things that would make it better :wink:
Marking stones and groups dead is currently a bit broken. If I click on any stone the whole board for that colour will be marked dead. SHIFT- and CTRL- clicking works better but then if I click on a group of stones it will only mark one stone dead.
It would be great if just normally clicking (without SHIFT or CTRL) it would mark that stone/whole group dead.
Also if you could click on an empty spot on the board to toggle where that point belongs to (black, white, neutral)
I haven’t been able to get the score estimation to automatically update on each new move. The score estimator itself will stay active but the score doesn’t seem to change accordingly. It would be great if one could just leave the score estimation turned ON and watch how the estimation progresses. Of course it would remember all marked dead groups and stuff like that :slight_smile: If a group that was earlier marked dead would connect to another group (edit) that was earlier NOT marked dead (end of edit) then it could reset the status for that group.

1 Like
4

I was just about to make a new thread, but @Pempu said everything I wanted to say. In particular:

5

As of 2015/09/24, simply clicking a group during estimation still marks the whole board for that colour dead.

6

Just curious:

The SE said black won by 0.5 point. However, when both players passed, the final SE says white won by a 0.5 point. If you click the SE, even now, it still says black won.

Are the SE during the game and the scoring after the game from two different AI’s?

7

Just curious:

https://online-go.com/game/32055139

The SE said black won by 0.5 point. However, when both players passed, the final SE says white won by a 0.5 point. If you click the SE, even now, it still says black won.

8

Are the SE during the game and the scoring after the game from two different AI’s?

9

That game you linked is non-existent. However, my guess is that it was anticipating some moves being made to seal off dame and possibly reduce territory that were not made. Score estimator is not the current score, but rather an anticipation based on a very basic monte carlo tree search.

1 Like
10

To be clear, I believe there are 3 score estimators:

  1. in game score estimator. This one is relatively basic, but can handle incomplete boards
  2. after game scoring tool. This one is also basic, but can possibly optimize based on the assumption that the board is finished (all boundaries are sealed). This one is the source of truth (but of course requires human assistance)
    3)The AI review score estimate. This is usually highly accurate, but you don’t get to see the internals. One thing that may throw it off is use of area scoring even if that was not the ruleset in the actual game. It may also differ from the scoring tool if the humans miss important life and death situations.
3 Likes
11

Wow, are you going for a necro record?

5 Likes
12

He can’t beat bugcat, to be honest.

4 Likes
13

I’m an idiot. That was totally unintentional. I was looking around for information on score estimator stuff and saw copacetic’s question.

Forgot I was in an old thread, and here we are!

3 Likes
14

A1

4 Likes
15

Cool old topic though.

I think you forgot at least one more, the widget that pops up if you click the “captures” area. What’s it called? It’s really helpful:

image

This one knows the scoring system (Chinese/Japanese/etc.) so it has the potential to be more accurate than the score estimator. But it has zero intelligence and absolutely no concept of dead stones. So you need to carefully seal off all boundaries AND kill off all dead groups with the analysis tool before you can use it.

4 Likes
16

Maybe that’s the end of game one?

2 Likes
17

Oh right! I used to use that when I first started on OGS and couldn’t find the estimator!

That’s a good one too!

I think the end-of-game tool can sort of handle unclosed boundaries, whereas the player card one does not.

I think it’s possible that the in game estimator (the “Estimate Score” button) could be the same as the end-of-game tool, but the wonky clicking to mark groups dead tells me it’s not.

1 Like
18

One thing that may throw [the AI review score estimate] off is use of area scoring even if that was not the ruleset in the actual game.

I have a suspicion that the scoring AI doesn’t just use area scoring, it uses specifically Chinese scoring, meaning that handicap stones don’t score. A pedantic point most of the time, but important in a close game.

I might be completely incorrect, though. Is this the case @anoek?

3 Likes
19

Accurate: Have score estimator respect ruleset · Issue #1274 · online-go/online-go.com · GitHub

EDIT: No wait, that issue is for the in-game estimator. AI review seems to be accurate. I think it either respects the score, or at least uses a ruleset that is functionally equivalent to the one being used.

2 Likes
20

Well, I am glad someone felt inspired to answer my ancient question.

Thanks!

2 Likes