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.

Sign up for GitHub

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

Jump to bottom

AMAF/RAVE implementation is too sloppy at the current point #2

Open
kalliduz opened this issue Feb 14, 2018 · 0 comments
Open

AMAF/RAVE implementation is too sloppy at the current point #2

kalliduz opened this issue Feb 14, 2018 · 0 comments
Assignees
@kalliduz
Labels
improvement

Comments

@kalliduz
Copy link
Owner

To estimate values quickly on big boards, it is needed to do a more or less dirty trick:

If you encounter a move on point X,Y for a player, and you do playouts for that move, let other moves on that point also profit from the assigned value by a certain amount.

This does of course increase the noise level for winrate-distributions over the board, but also it gives very fast access to potential exploration-worthy nodes.

The current implementation is only marginal and needs to be reworked, if not simply thrown out and rewritten. This is pretty important, since the speed of value estimations on a 19x19 board is far too slow to play against.
@kalliduz kalliduz self-assigned this Feb 14, 2018
kalliduz added a commit that referenced this issue Feb 17, 2018
@kalliduz
[REMOVED]
e7d8502 
- sloppy AMAF implementation REFS #2 @30m
- todo commentaries (outsorced as issues)
[ADDED]
- option to disable rating overlay when displaying the board
- generic tree structure REFS #8 @8H
kalliduz added a commit that referenced this issue Feb 25, 2018
@kalliduz
[ADDED]
3caa84b 
- additional display values (AMAFPlayouts)
- AMAF/RAVE implementation REFS #2 @4h
- adapted GameManager to minimally support playing with the UCTreeThread instead of old uct implementation REFS #8 @30m

[REMOVED]
- most playout heuristics (don't work too well with RAVE, bias to random playouts shouldn't be distributed among all nodes!)

[FIXED]
- some bug in UCT that caused the engine to select the global highest uct node instead of the best path (minimax) CLOSES #10 @2h

[CHANGED]
- some weights
- no more "random" montecarlo playouts needed, everything done now by the UCT
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@kalliduz kalliduz

Labels
improvement
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
1 participant
@kalliduz