Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

KataGo 1.1 #128

Open
alexpenson opened this issue Jun 18, 2019 · 3 comments
Open

KataGo 1.1 #128

alexpenson opened this issue Jun 18, 2019 · 3 comments

Comments

@alexpenson
Copy link

alexpenson commented Jun 18, 2019

The new release of KataGo promises to be a top-level engine with native support for high-handicaps and territory estimation. I'd love to try it out with goreviewpartner.
I believe there isn't yet a windows version, but it does have a version of lz-analyze. Let me know if I can be helpful.
Thanks!

https://github.com/lightvector/KataGo/releases/tag/v1.1

@ozymandias8
Copy link

ozymandias8 commented Jun 25, 2019

There is a windows build here:

lightvector/KataGo#2

I am also extremely interested in setting it up in GRP. I can get it to work as a GTP bot, but GTP bots cannot be used for analysis. I can't figure out any sort of workaround at present time.

Note: Windows version works in Sabaki using standard command line. I did have to install the c++ redist first.

gtp -model C:\KataGo\20b.txt -config C:\KataGo\configs\gtp_example.cfg

Edit: I wanted to add to your statement above. KataGo is also extremely beneficial in that the use of territory estimation means that it is useful for analyzing handicap games, or games with different komi. Leela Zero is extremely limited in this respect unfortunately. On the whole, KataGo is much more useful for analysis than LZ in my opinion.

Furthermore, lightvector chose to hard-code ladders into the neural net inputs. The Deepmind research shows that it takes a really long time for "traditional" zero bots to learn ladders. From a go-playing perspective, as opposed to a research mindset, simply hard-coding them seems better. This way we can get accurate ladder reading without very high visit counts.

@kennyfs
Copy link

kennyfs commented Nov 10, 2019

So, when will katago be useful for goreviewpartner?

@roy7
Copy link

roy7 commented Jan 26, 2020

Here's 1.3.1 with Windows binaries/etc.

https://github.com/lightvector/KataGo/releases/tag/v1.3.1

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

4 participants