How to count the game in a review? (also: how to use the estimator?)

I apologise for such a basic question, but I’ve been confused for the past half an hour, and don’t find an answer in the forum* (and I’ve not played on OGS for a while…).

I am doing a review of a game. I played a variation until the game is finished, with all dame filled. Is there any way to count the score, similarly to how it’s done normally at the end of a game?
The score estimator outputs “Black by 0.0” which doesn’t help at all (the komi is 5.5, so it cannot be the correct result). I am not looking for an estimate but for the actual score.

As a side remark, I am really confused by what is supposed to be happening when you click on the board in the score estimator. In my case, it seems to change the estimate randomly by plus or minus 0.5 point. Is there a explanation of this behaviour somewhere?

*Except for this thread in 2017, but things may have changed : How to count scores?

1 Like

In analysis mode you can click in the player card areas to get a very basic summary of the score:

So, adding up the points, this game shows a score of 27 to 31, or W+4. Note that it’s a completely dumb count assuming that all dame are filled and all dead stones are removed, and it won’t account for any special unusual scoring rules like territory in seki.

The post-game score estimator uses AI to evaluate the game and often comes up with slightly weird answers:

I don’t know where the 0.3 points of uncertainty comes from, but it’s probably something about AI never being totally sure that the game is done. Sometimes it’s even worse, especially if there is a ko left on the board.

2 Likes

Ah, thank you. That is well-hidden, and that is enough for me, i can add a few moves to capture the stones.

I am quite surprised that OGS would add an IA-based estimator and no way to actually count a game. To me, this is like the joke about the guy with a smartwatch with many functionalities but who cannot tell what time it is.

txwolf, I guess you suggest that I learn how to count. You didn’t have to answer me but there was no need to be rude like that.

2 Likes

If you fill in a few extra points it seems to correct to W+4.

My guess is because the stones aren’t completely uncapturable, in the sense that if White plays, Black passes, the three stones in the upper right might get captured.

That might be incredibly unlikely but it is one possible outcome which probably gets averaged changing the estimate.

I think normally I would imagine you can just round the score to the nearest integer or half integer depending on whether there’s a 0.5 in the komi.

Not quite sure about the OP one though.

1 Like

I just tried it in a review of mine and it works quite well.

From this variation:

I asked for the score evaluation and got this:

which isn’t perfect because of the black stones that aren’t completely dead.

Then I clicked on one of those black stones and got this:

which ins’t perfect yet because of a couple dame not yet filled.
When the situation is completely free of ambiguities, it should work well.

Could you share you review, so to check the actual case?

That should be done to mark dead stones when the SE doesn’t recognize them.
Or even the opposite: to mark live stones when the SE wrongly count them as dead.
In other words: click on a stone or group to change its life status.
Doing this in the wrong place could give weird results.

In my previous example I clicked on T16 stone, thus saying it’s dead.
Here is the result:

which is wrong, of course, but it comes from a wrong assumption (T16 dead).

3 Likes

I’m not really that surprised that there isn’t a counting phase in review mode of a game.

It doesn’t seem like it would be the most requested feature.

Previously a different estimator was being used before Katago and maybe that didn’t have funny fractional points.

Now a manual counting tool on the other hand does sound somewhat useful, and I think was discussed before but I can’t remember where.

I thought it was intended as: click on score to see the details:

It works also in the nearby area though.

Also not always the same!

Hmm… almost:

But you have an extra black stone… how is it? Wasn’t it white’s turn to play?

Going on:

I got it!!!

But if I go on… oh well!

1 Like

You can pass in analysis mode too.

Again the 3.9 is probably the very unlikely future that black loses the disconnected stones or something weird like that.

However, if that’s the only possible capture I’m kind of surprised it doesn’t estimate above W+4

I’m kind of guessing here, and of course in Japanese rules, it’d be a bit weirder, since you’d have to fill in an equal number of places each, if you really want the estimator to not do any fractional stuff. (It might still will it seems, but maybe it’ll be closer?)

Probably knowing to round it is ok at the late late stage of the game.

I can’t necessarily explain it

2 Likes

@Jhyn you open a thread with a title how to count … I send you a count video. you call me rude. i don’t know what’s wrong with you.

Since people asked, the review where the estimator gives me jigo with komi 5.5 is the following, move n°65 (of course this is not a game where a counting tool would be necessary)

I understand, I wasn’t clear enough. I meant what happens when you click on the board and there is no stone. Some blue squares appear and the scores changes, but the meaning isn’t clear to me. Let me screenshot an example (I click in the center of the board, score goes from W+42.7 to W+39.8. I click again, the squares disappear and score become W+46.2).

I’m not sure of what difference there would be between a counting phase and a counting tool. I would be happy with either of them.

I understand why an estimator can output a weird value when the game is finished; this is not its job. But it seems that it can make a mistake of more than half a point when the game is finished on a 9x9 (see above example), so it cannot replace a counting tool.

I think you didn’t read the first post or you are in bad faith. I don’t think there is anything to gain in this interaction, so let us agree to stop it here.

So this is still only my guess, but if I fill in different points from your variation

I can see W+0.5 which seems to match the 43 to 38+5.5 score.

While say in your variation, before testing what happens when white captures themselves, there’s maybe still the smallest possibility the black stones are captured with two moves at the X’s

I’m not arguing that this is “good” functionality or anything, I’m also just experimenting with it, and trying to explain what I’m seeing.

I think there’s still some weirdness though, that I’ve seen with a demo board for example where

So that’s true, that maybe a counting tool could be useful.

But it also does seem that in a late stage like this just clicking the scorecards as @Lys says

seems to work fine as a counting tool in simple cases.

In the example above it seems to update as I use the back and forward arrows, but it’s just not necessarily highlighting the areas or anything like the estimator.

2 Likes

Sure. It is a bit hidden, and you need to capture all the prisoners by hand for it to work, so it is a bit clunky, but that’s fine, I don’t use it everyday. So my probem is solved (thanks!) and it became a bit of an academic discussion.

I still find the situation weird; as in, estimating the score during a game is something a bit subtle that the player is supposed to do themselves, and counting at the end is a tedious task that I would like to automatise, but the interface seems to be focused on the opposite. This is what I tried to express with the metaphor of a smartwatch that cannot tell you the time.

2 Likes

That is intended to change the status of that intersection and the adjacent ones.
It’s useful when a dame is wrongly counted as territory, but it gives weird results when done on actual territory.

In my previous example the intersections at T14 and N11 are counted as white points while actually they should be filled in endgame.
It’s possible to turn off those white squares by clicking on them and changing them to blue.
I tried that, though, and it doesn’t work. The squares remains unchanged.
It was working in the previous version of the SE.
Now it only changes something when I click on unassigned intersections and does nothing useful. I think it’s a bug.

2 Likes

This is what it looks like to me, though I was wondering if I was missing some complicated-looking feature.
The change in the score is pretty jarring as well. Even if it is a fuzzy evaluation, you don’t expect it to change by 8 points without changing the status of some stones.
What do you recommend I do? Is there some kind of bug report I could fill?

1 Like

If you’re familiar with Github, that’s the place to file bug reports.
https://github.com/online-go

Otherwise the forum will work. You can use “Feedback/Bug Report” category for your message.

“When there are not enough playouts”
review: w+9.7
estimator: b+5.3

AI sensei recommends to play in top right and estimates w+7.3

but if I play below instead, its suddenly b+4.8