Position estimator

Sorry if this has already been asked or requested.

We have a scoring estimator which works reasonably well in my opinion. But, it is only really useful in the endgame and late middlegame.

How about optionally allowing people to use a position estimator? This would give you an idea of how likely it is that black wins provided that both players play near optimally starting from the current position. It could be implemented using Leela. Just feed the current board position to Leela and let her run a number of simulations on it.

I can understand that this might not be possible in real time because of a large number of overlapping requests from several ongoing games and the relatively heavy nature of the simulation operation. Maybe it could be allowed in correspondence games first.

It could say something like “Black win ratio at move 56 was 54.21%”.

Leela, andy many other top engines, already give win % values. If this were a good idea, you could have any engine evaluate the current position and return its assessment.

That being said, I think that a position estimator is a bad idea. If it were available in analysis mode, someone could check their ideas against it and go with whatever the computer says. Even if it weren’t, it’d be easy to abuse it by checking every move to spot any blunders by your opponent. Part of the game of Go is accurately evaluating the board position on your own. Offering a computer-based crutch cheapens that aspect of the game.

Honestly, I don’t even really like that there’s a score estimator. I’ll admit that I use it, from time to time, when I’m not sure what the balance of the game is, but I should really be required to count on my own. The fewer computer-based crutches we offer players, the better.

That being said, I wouldn’t have a problem with score/position estimators being available after the game. Finding mistakes and figuring out what happened during the game using all available tools is a helpful method for improving, but those sorts of tools should be restricted to post-game analysis, not used to boost winning chances during a game.

7 Likes

I think you should be able to turn off all of these features (scoring estimator, position estimator, analysis mode,…) if you want. That way there is no problem. If you don’t want to use the crutch, then disable it. However, the position estimator could really be useful in-game. For example, in teaching games.

There is always a possibility to cheat–install a strong go engine on your machine and let that one play for you, for example. I don’t think it is our job to limit options. Let the players do that. I, for example, almost always disable analysis mode in my games. Mostly because I don’t want to be tempted to use it myself. I don’t really care what the other person does. Their development is not my responsibility.

I do find your suggestion about position estimation afte the game nice. Actually, it would be really nice to just click a button and get a review created by Leela. We could for example provide this as a service for site supporters.

In terms of reviews, it’d be interesting to try to write a javascript-based version of LZ once the raw network has gotten a bit stronger. With optimal weights using the larger network sizes that should be coming soon (20x256 or larger), I’d imagine that you could write a relatively compact program for performing game analysis that could run on a user’s CPU without burdening the server.

For post-game analysis, just download the sgf and use Leela or whatever engine you want on your computer. No need to add a feature in OGS for that.

It might be timely to point out that this is COMPLETELY AGAINST THE RULES.

https://online-go.com/docs/terms-of-service :

No Cheating or Computer Help
You can NEVER use Go programs (Leela, Zen, etc.) or neural networks to analyze current ongoing games unless specifically permitted (e.g., a computer tournament).

So if you are doing this yourself, you are cheating.

7 Likes

Thank you GreenAsJade! I am not.

1 Like

People who want to cheat, will cheat. There is nothing to stop them except the honor system.

That being said, a button to see the win % right in the game itself is such a big convenience that those who normally do not cheat may be tempted to do so.

What I was pointing out is that there is no point in proposing a mechanism in OGS to do something that is against the rules.

(At the same time, I was pointing out what the rules actually are, of course)

2 Likes

I’m curious, where is the line drawn with external software assistance?
I have used the KGS client (cgoban) before to get a quick count of the current position, so I might be guilty to a certain degree. But in my eyes it’s a more accurate (maybe assuming is a better word for it) version of the score estimator already available on OGS and it didn’t assist me any further than that.

(For reference, I started out on DGS where the rules say that everything is fair game as long as you’re not getting suggestions for next moves from an external source, so I started playing here with a similar mentality)

There are no more words than the ones I quoted, so it’d be down to personal judgement. No-one can prosecute anyhow - it’s an honour system.

“Go Programs” are prohibited. I think that broadly, one would expect a “Go Program” to be a program that determines moves, or estimates win percentages (since Leela is given as a Go Program, and Leela does it).

Personally, I think that since OGS has a score estimator, score estimators are not considered as “Go Programs”, so whether you use KGS or OGS score estimator is neither here nor there.

GaJ

My personal opinion is that even the current score estimator should be against the rules. Identifying which areas are secured territory, which are still open, counting and so on should be done by the player without any assistance, otherwise we are already playing a different game than if you play go on a real board.

And whilst I know that the score estimator is exactly that, an estimator, I have found that by repeatedly clicking it and seeing which areas the estimator consistently counts as territory for one side, I can get information that I would otherwise have been far less sure of.

And yes, I admit it to using the feature :slight_smile: It’s a bad habit I’ve been meaning to stop.

6 Likes

Agree with Stephen.

I can turn off Analysis for both players, which is a crutch compared to real Board Go, before a game—but that disables Conditional Moves, which slows Correspondence games significantly. (No need to explain to me how they’re necessary to be linked in codex; I’m just pointing out this inconvenience of having to disable both).

I can’t turn off the crutch of Score Estimator for both players—sincerely wish I could.

I’ll explain that anyway as a side note:

It’s not that analysis and conditional moves are linked codewise. The reason why disabling analysis also disables conditional moves is that you can use conditional moves feature’s interface to analyze move trees.

2 Likes

I agree with StephenC20XX. It should be possible to disable the scoring estimator. For those who are trying to learn proper counting, it can just be too much of a temptation to click the scoring estimator.

That being said, I think the key is that you should be able to disable it. You don’t need to disable it for everyone. I think the same would apply to a position estimator.

1 Like

Me three. I totally hate how much I use that thing.

It is really astounding how much my reading improves when I play a round of “Fast Correspondence Tourney”, where analysis is turned off. It improves over about the first 2 weeks of play: first week is a headache, then second week gets to “wow, I don’t even reach for the button”, then it’s really pleasant after that to be actually reading.

As soon as the tourney is over and I hit the first tricky situation: reach for the button again.

I wish analysis were turned off in Ladders.

And for the same reason I wish the score estimator were turned off.

1 Like

I think analysis and any other tools of that nature should be turned off by default.

4 Likes

This topic was automatically closed 91 days after the last reply. New replies are no longer allowed.