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

Another criterion of settledness #215

Closed
kaorahi opened this issue Oct 10, 2020 · 6 comments
Closed

Another criterion of settledness #215

kaorahi opened this issue Oct 10, 2020 · 6 comments

Comments

@kaorahi
Copy link
Contributor

kaorahi commented Oct 10, 2020

I post this comment since I noticed the new AI strategy "Simple Style" in v1.5.1 branch.

As an indicator of the board atmosphere in lizgoban, I have been using U = \sum (1 - abs(ownership(i,j))) over all coordinates (i,j) of "existing stones" on the board. The red dots in the screenshot indicate progress of U through a game. In my impression, this U well matches my feeling of unsettledness.

unsettled

  • It increases from the opening to the middlegame, and decreases in the endgame.
  • It increases in large battles, especially in ko fights, and decreases when stones are clearly alive or dead.
  • It tends to be large in KataGo's handicap games because of many large Miai.
  • It tends to be large in recent professional games and small in old Japanese championships before 1990.

How about using this criterion instead of the current definition of settledness in "Simple Style"? I am also interested in a berserker style that tries to maximize U.

@sanderland
Copy link
Owner

Yes it is very early testing, your idea is interesting, but it doesn't seem like a very 'strong' criterion since if the only effect is on the last stone played, the difference will be quite small.
I may try a few different variants and make a pre-release where people can test them for me.

https://online-go.com/player/768720/katrain-balanced has some experiments with 'berserker' mode and it is scary.

@kaorahi
Copy link
Contributor Author

kaorahi commented Oct 12, 2020

thx!

Expecting a berserker or a chaos lover, I built katago b9ce89c3 in the master branch, modified katrain/core/constants.py in 2bf1846 as

-    "settled_weight": [x / 4 for x in range(0, 17)],
+    "settled_weight": [x / 4 for x in range(-20, 17)],

and tried ai:settle on settled_weight=-5, opponent_fac=1, attachment_penalty=0, tenuki_penalty=0 with max_points_lost=5, min_visits=3, wide_root_noise=0.1. But AI's moves seemed quite regular. I may need to increase max_time as my machine is slow.

@sanderland
Copy link
Owner

Note that you don't need to edit constants.py to go outside the range, the text box accepts pretty much anything. Low visits could be an issue, yes.

@sanderland
Copy link
Owner

This is what happens for me on those settings: https://gokibitz.com/kifu/ByB0QAWwD

@sanderland
Copy link
Owner

Deprecated due to lack of feedback

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

2 participants