Search found 171 matches

by dkappe
Sun Feb 10, 2019 1:19 am
Forum: Computer Chess Club: General Topics
Topic: Leela0 (LC0) cpu Initial Observations
Replies: 13
Views: 1552

Re: Leela0 (LC0) cpu Initial Observations

supersharp77 wrote:
Sun Feb 10, 2019 12:53 am
"Put up or shut up time draws near"..... :D :wink:
Not sure what that means. Nothing draws near. Every month or so new features are added and newer, stronger nets come along.

If folks are running misconfigured lc0 or lc0 on cpu, I’m more than happy to help.
by dkappe
Sat Feb 09, 2019 4:00 am
Forum: Computer Chess Club: General Topics
Topic: Leela0 (LC0) cpu Initial Observations
Replies: 13
Views: 1552

Re: Leela0 (LC0) cpu Initial Observations

On my gpu (a 1070) the 256x20 nets do about 4000 nodes per second. On a cpu, we’re talking 30-40 nodes per second. The smaller 112x9-se net will do many 1000’s of nodes per second on cpu. It plays about 300 elo stronger than a full sized net on cpu. (Remember, the net weights are the pb.gz files.) I...
by dkappe
Sat Feb 09, 2019 2:19 am
Forum: Computer Chess Club: General Topics
Topic: Leela0 (LC0) cpu Initial Observations
Replies: 13
Views: 1552

Re: Leela0 (LC0) cpu Initial Observations

Can I ask which network files you are using? If you are using a full size network on cpu, it will do quite poorly. Give some of my distilled networks a try. The 112x9 network is especially good on cpu. I’d estimate it as about 2800-2900 ccrl. https://github.com/dkappe/leela-chess-weights/wiki/Distil...
by dkappe
Fri Feb 08, 2019 9:40 pm
Forum: Computer Chess Club: General Topics
Topic: TCEC 14 final: Lc0 just lost 3 games with tactical..
Replies: 32
Views: 4723

Re: TCEC 14 final: Lc0 just lost 3 games with tactical..

They’re from a variety of sources, including tcec and cccc
by dkappe
Fri Feb 08, 2019 9:17 pm
Forum: Computer Chess Club: General Topics
Topic: TCEC 14 final: Lc0 just lost 3 games with tactical..
Replies: 32
Views: 4723

Re: TCEC 14 final: Lc0 just lost 3 games with tactical..

fwiw, i find t40 passing all of these (these being the 3 starkest blunders i've found so far) & avoiding a couple other, less-losing moves i've tested, but have only analyzed the Kf1 blunder at several MNs. 1r1q1rk1/p1pbbp2/7p/8/2PPnP2/1P6/PB2B2p/RN1Q1R1K w - - 0 20 31 [bm Qd3 (??)] 20. Rf3?? Bg4! ...
by dkappe
Fri Feb 08, 2019 8:22 pm
Forum: Computer Chess Club: General Topics
Topic: TCEC 14 final: Lc0 just lost 3 games with tactical..
Replies: 32
Views: 4723

Re: TCEC 14 final: Lc0 just lost 3 games with tactical..

Deep Blunder rides again. There are probably some ways of improving search in the short term, maybe with a clever use of a complementary ab engine. In the long term, training will have to be revisited. When training Ender, I would look where sf10 100k node search (converted to the range -1.0 to 1.0)...
by dkappe
Wed Feb 06, 2019 1:49 am
Forum: Computer Chess Club: General Topics
Topic: Test40 late nets scale better than test10 and test30 nets to longer TC?
Replies: 7
Views: 1552

Re: Test40 late nets scale better than test10 and test30 nets to longer TC?

Laskos wrote:
Tue Feb 05, 2019 7:56 pm
I checked in an earlier thread the scaling of test10 nets, coming to conclusion that at longer TC and high Leela Ratio, they do not scale very well against SF10:
I have put the t10, t30 and t40 nets on a scale and they all weigh the same. One can conclude that they all scale the same.
by dkappe
Wed Feb 06, 2019 12:45 am
Forum: Computer Chess Club: General Topics
Topic: Lc0 ... the GOOD and the BAD
Replies: 54
Views: 7678

Re: Lc0 ... the GOOD and the BAD

You guys have been focusing on obvious tactical blunders. I've been focusing on positional blunders and wide swaths where engines in general and Leela in particular have no positional understanding and play the position either aimlessly (sometimes for several moves) or make a single positional move...
by dkappe
Tue Feb 05, 2019 2:03 am
Forum: Computer Chess Club: General Topics
Topic: Lc0 ... the GOOD and the BAD
Replies: 54
Views: 7678

Re: Lc0 ... the GOOD and the BAD

One wonders how AlphaZero was able to overcome such weaknesses that Leela still has -- occasional tactical blindness, and relatively weaker and sloppier endgame play. Was it just a matter of a larger amount of hardware resources allocated to training, or perhaps a smarter approach/methodology? I do...
by dkappe
Mon Feb 04, 2019 8:21 pm
Forum: Computer Chess Club: General Topics
Topic: Lc0 ... the GOOD and the BAD
Replies: 54
Views: 7678

Re: Lc0 ... the GOOD and the BAD

There are a few things that might help. One of them is basic certainty propagation (MCTS Solver) that will help in tactical and endgame positions and overall in self-play. Should be merged soon into an upcoming version of lc0. Now that is very interesting and seems that it would totally fix the rar...