The poor evaluation function of lc0

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

Moderators: hgm, Rebel, chrisw

Uri Blass
Posts: 10315
Joined: Thu Mar 09, 2006 12:37 am
Location: Tel-Aviv Israel

The poor evaluation function of lc0

Post by Uri Blass »

[d]8/8/4P3/5k2/3B4/5pP1/6r1/5K2 b - - 2 62

In the bonus games lc0 played 62...Kxe6 against stockfish8 with evaluation of -12.44
syzygy
Posts: 5566
Joined: Tue Feb 28, 2012 11:56 pm

Re: The poor evaluation function of lc0

Post by syzygy »

It might not be able to draw this with white, so -12.44 might be accurate from that point of view.
zullil
Posts: 6442
Joined: Tue Jan 09, 2007 12:31 am
Location: PA USA
Full name: Louis Zulli

Re: The poor evaluation function of lc0

Post by zullil »

Uri Blass wrote: Sat Aug 18, 2018 11:48 am [d]8/8/4P3/5k2/3B4/5pP1/6r1/5K2 b - - 2 62

In the bonus games lc0 played 62...Kxe6 against stockfish8 with evaluation of -12.44
Well, Sf-dev would also play Kxe6, with evaluation -5.65. At least, without endgame tables.

info depth 78 seldepth 90 multipv 1 score cp 565 nodes 554621123 nps 3729473 hashfull 999 tbhits 0 time 148713 pv f5e6 d4b6 g2g3 f1f2 g3h3 b6c7 e6f5 c7b8 f5e4 b8a7 h3h6 f2f1 h6h2 a7c5 h2h3 c5a7 h3h6 a7c5 h6a6 f1f2 a6a2 f2f1 e4f5 c5e3 a2c2 e3a7 c2c1 f1f2 f5e4 a7b6 c1c2 f2f1 e4e5 b6a7 e5f5 a7e3 f5g4 e3a7 c2c1 f1f2 c1d1 a7b6 d1h1 b6c7 h1a1 c7b6 a1a2 f2f1 a2g2 b6c5 g4g3 c5d6 g3h3 d6e5 g2e2 e5d6 e2d2 d6b8 d2d1 f1f2 h3g4 b8a7 d1d2 f2f1 g4g3 a7b8 g3h3 b8c7 d2c2 c7f4 c2b2 f4e5 b2g2 e5d6 g2e2 d6b8 h3g4 b8a7 g4f4 f1g1 e2g2 g1f1 f4g3 a7b8 g3h4 b8d6 h4g5 d6e5 g5f5 e5d4
megamau
Posts: 37
Joined: Wed Feb 10, 2016 6:20 am
Location: Singapore

Re: The poor evaluation function of lc0

Post by megamau »

It is very "human like".
If you don't know it is a draw (no tablebase / not knowing the ending) then you would think black has a big advantage.