Confused about how "estimate score" works

Hello,

In this game, Black won by 5.5 points. However, near the end of the game (or indeed right now), using Estimate Score said that Black was winning by 12.5 points. There’s no ambiguity, as the game is over, so I’d expect those numbers to be the same. (And because it’s a 19x19 game, I really don’t feel like counting it up manually to see what the real score is.)

Why are these scores different, and which one is correct?

Thanks!

1 Like

Komi in your game was 0.5. I’m only guessing, but maybe “estimate score” automatically assumes 7.5 komi for AGA/Chinese rules regardless of the komi actually used?

Interesting idea, but the math doesn’t check out.

If SE “assumed” komi when there wasn’t any, white would have won the game according to SE (5.5 - 7 = -1.5). However, SE’s margin of victory for b is larger than the actual margin. Considering that SE is wrong most of the time, no one should take it seriously. It’s only a concern if the scoring algorithm is wrong, which happens from time to time - counting dame as points and vice versa.

1 Like

Lol, oops :slight_smile:

@Glyconis this happened because there was a handicap. black has more stones on the board than white, which SE counts as points.

7 Likes

There we go :slight_smile:

game estimate uses chinese rule but game uses Japanese rule. if there was a ko during the game the score may be off by 1 point.

It had occur to me twice that using estimate I won by 0.5 and when the game finalize I lost by 0.5. :unamused:

1 Like

Guess estimate can be off by much more than just 1 point. That’s because its a guess.

It has to guess what might be played, and it has to do that very quickly, so it doesn’t try many options.

It’s at least +/-5 even quite close to the end.

It’s no substitute for careful counting if things are close. It’s just a rough estimate.

There’s no point consulting estimate score. It’s not capable of making accurate or even plausible judgements until the game is over, at which point OGS will tell you the score anyway.

Of course it is prone to error during mid game, or even close to end of game.

The story I am referring to is actual end of game where every space that should be filled is filled and there is no ambiguity. But due to the fact that there was a Ko during the game the rules count differently.

2 Likes

Recently I discovered that score estimator may be wrong.
Here is a topic where I discussed my doubts:

Look at the pictures I posted in it.

The fact is that on the roadmap to be a full go player, counting score is very important. Maybe not urgent at first, but very important indeed.

Now I try to count as frequently as possible to train myself and get a quite good counting in a short time.

2 Likes

I counted you game, but I don’t know AGA rules (that apply in your game) so I counted Japanese:

white has 99 territory +7 captures + 0,5 komi=106.5
black has 110 territory + 2 captures = 112
black wins by 5.5

At move 175 (just before both players passed), score estimator says “black by 12.5”, which is wrong.

It’s at least +/-5 even quite close to the end.

But I’m not talking about “close to the end”. I’m talking about the literal end of the game.

As was mentioned, it looks like it’s because SE was mistakenly counting handicap stones as points (which it is not supposed to do?), while the actual scorer does not. So it’s a bug.

Score estimator always counts in Chinese rules. (Don’t ask me why, it is some programmer thing where Chinese rules are easier to preditc before the end of the game. Also why all bots play in Chinese rules.) Otherwise we would need a separate estimator for every rulesets, which would be crazy.

In Chinese rules not only surrounded empty intersections, but every stone as well are count as points.
So as far as the estimator is concerned, it counted correctly, only it counted under different ruleset, which may sometimes change the result. Which is unfortunate, but to be honest with you it is very improbable at the moment that an estimator for every ruleset gets implemented anytime soon. :frowning:

What do you mean by progamator? Did you make a typo?

yes :frowning:

Chinese rules compensate white by 1 point for every handicap stone, so the score estimator is still wrong. The score estimator counts number of stones + amount of territory + komi.

Using the current algorithm, black has 78+109+0, white has 85+89+0.5. This gives 12.5.

The algorithm should be stones + territory + komi + (handicap given - 1). White would get an additional 7 points, since they gave 8 handicap stones, which would give an estimate of black by 5.5 points.

2 Likes

And you are of course correct. Silly me should have checked properly :frowning:
I will file a bug report

1 Like

Hi guys,

I actually would like to know how it works. Not at low level of detail, but the main functionality. When I click in the different areas the scoring changes randomly.

What it supposed to do? (when you click on different parts of the board)

I was expecting I would be able to click on squares and change them from black to white and the other way around. I used to play Go online in an app and it used to work that way. Also if you click on a group of stones it changes it from dead to alive, and the other way around.

1 Like

Hello - basically the “click to change the estimate” function is broken right now, sorry. Our developer is working on a new system for scoring and estimating and decided to rather work fully on the new one instead of trying to fix the old.

And welcome to the forums :slight_smile:

3 Likes