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

Unlimited time #85

Closed
kennyfs opened this issue Aug 27, 2018 · 4 comments
Closed

Unlimited time #85

kennyfs opened this issue Aug 27, 2018 · 4 comments

Comments

@kennyfs
Copy link

kennyfs commented Aug 27, 2018

You can use this command to set an infinite time.
time_settings 0 1 0
Why are you not using it?
It is useless to adjust visit because there is a time limit.

@pnprog
Copy link
Owner

pnprog commented Sep 2, 2018

Hi!

GRP does not support unlimited time, and does not plan to for the future. The purpose of GRP is to automatize the analysis to be done without human supervision, so it goes against unlimited thinking time (that require manual operation).

Most of the bots (Ray, AQ, and currently under implementation Pachi) does not support the feature but you can use Leela with the original GUI or Leela Zero with the Lizzie software for real time analysis and unlimited thinking time.

For Leela and Leela Zero, in GRP Settings, you can use the fields Time per move (s) to use the time_settings command. So a time per move of 25 s means GRP will use the command time_settings 0 25 1 and Leela/LZ should use only 25s per move. So if you decide to use that parameter, you can remove the --playouts parameter from the command line.
If you let this setting empty (or set it at zero) GRP does not use the time_settings command, so Leela/LZ will decide only based on the command line parameter.

@Ishinoshita
Copy link

@pnprog With LeelaZeron engine, with:

  • Parameter --gtp --noponder --weights LZ202 -v 200
    -Time per move (s) 120

when I do Test/genmove black, I get:
Using 4 thread(s).
...
Thinking at most 27.3 seconds...
N eval=0.446980
Playouts: 0, Win: 4.47%, PV:
Playouts: 1, Win: 8.15%, PV: Q4
Playouts: 3, Win: 13.79%, PV: Q4 D16
...
Playouts: 39, Win: 36.38%, PV: D16 Q4 Q16 D4 R6 R14
D16 -> 15 (V: 44.68%) (N: 19.33%) PV: D16 Q4 R3 Q3 R4
Q4 -> 9 (V: 44.61%) (N: 19.80%) PV: Q4 D16 C17 C16
Q16 -> 9 (V: 44.48%) (N: 19.17%) PV: Q16 D4 C3 C4
D4 -> 9 (V: 44.46%) (N: 19.68%) PV: D4 Q16 R17 R16
4.3 average depth, 7 max depth
27 non leaf nodes, 1.56 average children
43 visits, 15422 nodes, 42 playouts, 1 n/s

Same when I leave Time per move (s) blank.
Does this test means that the analysis will be truncated after 27s (I use CPU only, so cannot get 200 visits in less than 30s most of the time) ?

@Ishinoshita
Copy link

Forget it. I've launched my first sgf review and found that the variations # sum-up to to the requested visits parameter.
And thank you for GRP. Fantastic tool !

@pnprog
Copy link
Owner

pnprog commented Feb 3, 2019

Leela and Leela Zero can be very stubborn regarding the time spend to think on a move.
My advise would be to put both a time limit and a playout limit. For examples:

  • You want a fixed time of 45 seconds, then you can put a time per move at 45s, and use --playouts 1000000
  • You want a fixed number of 5000 playouts, then you can put a time per move at 10000s, and use --playouts 5000

Leela(Z) will normally stop when either the time limit is reached, or the playouts limit is reached.

@pnprog pnprog closed this as completed Feb 3, 2019
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

3 participants