It is currently Sat Oct 23, 2021 7:07 pm

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 25 posts ]  Go to page Previous  1, 2
Author Message
Offline
 Post subject: Re: Trying to Modernize Sensei's Library
Post #21 Posted: Tue Aug 17, 2021 6:53 am 
Dies with sente

Posts: 80
Liked others: 135
Was liked: 2
OGS: Elom hmW K
Universal go server handle: Elom, Windnwater
Kirby wrote:
Personally, I am in favor of maintaining the pages for "outdated" joseki; maybe you could make a new page for the AI version of a particular pattern.

Senseis library serves various purposes, and I find value in having the old explanations of patterns that may not be AI-approved. For a page explaining a pro's thoughts on a joseki, I'm not really a fan of seeing the win rates, etc., thrown up there.

If I want to know what an AI thinks, I can open up an AI and give it the position I'm interested in analyzing.

The value from senseis library, imo, is different; you get a human perspective, as well as historical context behind why pros thought certain moves were good and bad. Maybe the page is inaccurate or gives suboptimal moves. But I'm not looking for an AI analysis when I'm going to senseis.

Basically, I think there's some element that's lost when updating these pages and just copying the stats you get from your GPU. We don't need senseis library for this type of analysis.

---

I know some people disagree with my opinion, and want to run KataGo on whatever position that's documented on the internet.. That's why I'd recommend making separate pages; AI josekis for those interested in the AI analysis, and the old ones for those interested in the historical context. There's at least some people interested in the latter (me, for example).


I really think it would be best to have both. Having only AI opinion or pro opinion wouldn't feel correct. Joseki: by definition it means a sequence pros thought there were good reasons to believe it was equal. They may or may not think it's equal now, but the point. I'd like to see at the top of the page the winrates, a short section on AI opninion (it doesn't need to be long since different computers are different and the time spent by the bot evaluating each move is different and they update over time so . . . there's no meaningful single winrate update for any bot), and then move swiftly onto pro opinions on the sequence from a variety of pros. It would need to be both in that proportion for it to feel authentic and wholesome for me.

_________________
Practical honeytarian

Top
 Profile  
 
Offline
 Post subject: Re: Trying to Modernize Sensei's Library
Post #22 Posted: Tue Aug 17, 2021 6:59 am 
Dies with sente

Posts: 80
Liked others: 135
Was liked: 2
OGS: Elom hmW K
Universal go server handle: Elom, Windnwater
Kirby wrote:
I like win rate in doing private analysis on games, but I don't like using it as any sort of authoritative source, since win rate for a position can vary:
* What engine are you using?
* How may blocks in the network?
* How fast is your GPU? How long have you let the AI calculate moves?
* What is komi set to, and what is the ruleset?
* How well does the computer's win rate correlate to the realistic win rate for a human? (e.g. something with a lower win rate may be more playable for humans)

It's not that I don't think that the win rates are useful. It's just that I think of win rates in a more fluid sense, in that the exact numbers can vary based on a number of variables. As such, I typically prefer to investigate win rates interactively with software, rather than assuming a static value that was retrieved from one particular version of one particular AI based on one particular number of play outs.

That being said, I may have an old fashioned view on this. Or maybe I'm simply missing the pre-AI era ;-)


Interestingly I strongly believe that the AI section on a joseki page should be short and sweet (given the possible factors that affect winrate it can't be all that useful except hen you're using your own or a familiar bot(s) you know and understand) . . . Yet I also believe just as strongly that it should indeed be there and be first. It seems contradictory, but perhaps these strong contradictory intuitions are signs that that exact balance is the one to go for . . .

_________________
Practical honeytarian

Top
 Profile  
 
Offline
 Post subject: Re: Trying to Modernize Sensei's Library
Post #23 Posted: Tue Aug 17, 2021 12:01 pm 
Beginner

Posts: 11
Liked others: 0
Was liked: 6
Rank: OGS 4 kyu
OGS: yuzukitea
I wrote this page: 4-4 Point Traditional Slide Joseki, 2-8 Invasion

I have mixed feelings about how AI joseki should be presented. I agree with others that it's not very helpful to present an AI score of a joseki played out on an empty board (or even an invented arbitrary board position), and it can produce a misleading understanding about what AI thinks about a certain variation.

I think it's definitely better to assess positions in professional games with AI. I've been doing this with a few professional games with every variation, and I feel like I'm fairly comfortable with saying "AI doesn't like this traditional variation" in a hand-wavy sense (i.e. AI views the extend joseki linked above as -1 points in the context of the professional games I checked), but I wouldn't really feel comfortable throwing a number onto it without being able to do statistics running many more professional games through AI.

Even then, I don't really consider myself authoritative enough to say that "X joseki is bad". There's sort of a catch 22 of professionals no longer play this joseki ever since 2016 (so there's no hits in Waltheri after 2016), but at the same time I don't have a documented source from any professional saying that "X joseki is bad" or explaining why "X joseki is bad". It's ultimately just inference and guesswork based on the lack of any professional play.

Meanwhile there are strong amateurs like Badukdoctor saying the extend variation is "good" and there's likely a great number of older joseki books still saying that the extend is "good".

I just don't think I'm qualified enough to provide a good personal opinion, so the best I can do is present both.

Top
 Profile  
 
Offline
 Post subject: Re: Trying to Modernize Sensei's Library
Post #24 Posted: Thu Oct 07, 2021 1:26 pm 
Dies in gote

Posts: 32
Liked others: 1
Was liked: 9
Rank: OGS 8k
DGS: bugcat
I wonder whether anyone would like to help update Sumire's article, which hasn't been edited since May 2019.

https://senseis.xmp.net/?NakamuraSumire
https://senseis.xmp.net/?topic=11193 (discussion of updating)

And also Joanne Missingham's, which hasn't had a content update since 2012, apart from noting her 2015 rank up to 7p.

https://senseis.xmp.net/?JoanneMissingham
https://senseis.xmp.net/?topic=11206 (discussion of updating)

Top
 Profile  
 
Offline
 Post subject: Re: Trying to Modernize Sensei's Library
Post #25 Posted: Mon Oct 11, 2021 7:28 am 
Dies in gote

Posts: 32
Liked others: 1
Was liked: 9
Rank: OGS 8k
DGS: bugcat
If anyone would like to start articles on the Yunguseng Dojang or AwesomeBaduk, that'd be appreciated.

We already have a short one on the Nordic Go Dojo and now a longer one on the Osaka Go School.

No-one has yet come forward to help update the Sumire or Joanne Missingham articles.

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 25 posts ]  Go to page Previous  1, 2

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: No registered users and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group