Lc0 defeats SF in a position where SF think it stand better.

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
User avatar
Nordlandia
Posts: 2471
Joined: Fri Sep 25, 2015 7:38 pm
Location: Sortland, Norway

Lc0 defeats SF in a position where SF think it stand better.

Post by Nordlandia » Tue Dec 03, 2019 8:14 am

Looks like Lyudmil Tsvetkov was right. He shared this FEN couple of years ago, claiming white was winning.




Spliffjiffer
Posts: 226
Joined: Thu Aug 02, 2012 5:48 pm
Location: Germany

Re: Lc0 defeats SF in a position where SF think it stand better.

Post by Spliffjiffer » Tue Dec 03, 2019 3:35 pm

33...Qd8 is loosing imo, what does Leela think about 33...Qd7 !? instead...at least with SF-Bluefish i couldnt find anything that promised a positive score for white (0.00)...manual investigation time: around 1 hour on a slow PC
Wahrheiten sind Illusionen von denen wir aber vergessen haben dass sie welche sind.

zullil
Posts: 5680
Joined: Mon Jan 08, 2007 11:31 pm
Location: PA USA
Full name: Louis Zulli

Re: Lc0 defeats SF in a position where SF think it stand better.

Post by zullil » Tue Dec 03, 2019 10:53 pm

Spliffjiffer wrote:
Tue Dec 03, 2019 3:35 pm
33...Qd8 is loosing imo, what does Leela think about 33...Qd7 !? instead...at least with SF-Bluefish i couldnt find anything that promised a positive score for white (0.00)...manual investigation time: around 1 hour on a slow PC
Lc0 (network J13B.3-200) after 100 M nodes certainly considers 33...Qd7 to be best:

info depth 19 seldepth 64 time 1716840 nodes 100000113 score cp -89 wdl 173 351 476 hashfull 107 nps 58247 tbhits 0 multipv 1 pv e8d7 e3f2 d7g7 f1d1 f8f7 f2g3 c8e8 c2e4 h7h6 f3d3 f7d7 g1h2 e8f8 d1f1 d7f7 f1d1 f7d7 d1d2 f8f7 d3d4 f7f8 e4c2 g7g8 d2f2 b7c8 c2e4 c8b7 f2f3 g8g7 e4c2 d7f7 c2b3 f8e8 f3d3 f7d7 b3d1 e8f8

17.3% chance that Black wins, 35.1% chance of a draw, 47.6% chance that Black loses.

33...Qd8 is evaluated as clearly worse for Black:

info depth 19 seldepth 64 time 1716840 nodes 100000113 score cp -146 wdl 102 298 600 hashfull 107 nps 58247 tbhits 0 multipv 5 pv e8d8 c2b3 d8d7 e3f2 c8e8 f2h4 b7c8 h4f6 d7a7 f1c1 a7b7 g1h2 c8d7 g2g4 b5b4 f4f5 b7b5 c1c2 b5a5 f5e6 e8e6

Post Reply