TalkChess.com

LS14 that Pohl tested 46 elo stronger than LS13 after go 50M nodes. info depth 37 seldepth 90 time 1772501 nodes 27383748 score cp 2597 hashfull 1000 nps 15142 tbhits 0 multipv 1 pv f1g2 c8g4 c2c4 c7c6 d1b3 e7e6 b3b7 b8d7 b1c3 g8e7 c4d5 e6d5 d2d4 a8b8 b7a6 b8b6 a6d3 e7g6 h2h3 g4e6 g1f3 f8d6 h3h4 h7...

Fen after 60. Nd3 8/2k5/1p6/pP6/P7/2PN1B1p/4K3/1r6 b - - 9 60 Redfish spiked to 5.3 here, but here's analysis After 60...Ra1 info depth 78 seldepth 98 multipv 1 score cp -14871 nodes 1444012721031 nps 34440357 hashfull 1000 tbhits 13704844823 time 41927925 pv d3f4 h3h2 f4d5 c7b7 c3c4 b7a7 d5c3 a1g1 ...

Cool! Hopefully we'll see what kiudee tuner can do with it sometime too!

Leelenstein is defending its title, and in the lead at WCEC https://www.twitch.tv/navratil25 Its 2 RTX gpus with a strong I9. Leela ratio is a little worse than last TCEC season. If he disabled a few threads from SF it matched TCEC's ratio, but this tournament is at full bore. You can see a lot more...

I debated over using q output that goes from -100 to 100 but decided to use winrate percentage that goes from 0 to 100 instead, since I think it's just a little bit more clear - and the CP output doesn't really mean anything with every engine using a different formula and it being so far removed fro...

The threshold for a won position is probably about 70% or perhaps slightly higher, 71 or 72%, based on everything I've observed. For 1 g4?, it is about 68% for Black with 1...d5, so probably not a forced win although close enough to the line to leave that as a possibility. Leelenstein 13 already is...

Using the kiudee settings Pohl used? WCEC is using these similar settings: "Threads": 3, "MinibatchSize": 512, "Cpuct": 1.9000 "NNCacheSize": 20000000, "maxCollisionEvents": 32 "CPuctBase": 15000, "CPuctFactor": 2.817, "TradePenalty": 0.0 "TradePenalty2": 0.0 "TempDecayMoves": 0, "TempCutoffMove": 2...

pohl4711 wrote: ↑

Tue Mar 10, 2020 11:29 am

NN-testrun of Lc0 0.23.3k Leelenstein 14 finished - a fantastic result! Massive progress to Leelenstein 13 (+46 Elo) and the clear world's number 1. Congrats to Jjosh - great job.

https://www.sp-cc.de

Cool! We'll see how it performs at WCEC and CCC

I'm hoping that it's possible for Leela CPU to train from a large PGN file instead of selfplay. If so, does anyone know how to explain it simply enough for an average Joe to understand it? There are many different levels to do this. If you just use straight PGN it creates a very weak policy head (l...

I wonder what would happen if the entire Lichess database was used to train a neural network. It wouldn't play very strongly but could possibly play in a way indistinguishable from humans. That is what the Darkqueen net is actually. Although latest ones have added a little bit of other data as well.