Page 1 of 1

Knights at g5 h5 by lc0

Posted: Fri Nov 09, 2018 11:32 pm
by Javier Ros
I have detected a curious maneuver of knights in g5 and h5 by lc0 in two games of a test that I am doing. In both cases the maneuver leds to a winning attack.

The test compares lc0 with AlphaZero against Stockfish 8 without opening book.
Lc0 version 0.19 rc2 with net 11261 on Asus Geforce GTX 1060 3 Gb tc=1500+10 (secs) against Stockfish 8 on i7 4790S 6 logical cores tc=710+4.73 (secs)

with LeR=1, ponder Off, Hash tables 1 Gb. under Cutechess cli. Turbo Boost disabled to avoid speed variations due to other processes.

In the first game, the evolution of the knight throughout the game is very beautiful and devastating.










Re: Knights at g5 h5 by lc0

Posted: Sat Nov 10, 2018 8:45 am
by Javier Ros
The same at g4 h4!! at move 34
and knights in h4 h5 at move 20






Re: Knights at g5 h5 by lc0

Posted: Sat Nov 10, 2018 11:53 am
by Werewolf
I have long held a theory in my own personal over-the-board human chess which I haven't seen in a any chess book: that there is an advantage of having two knights side by side, especially if near the opponent's king.

Knights often make each other redundant, but when they are side by side, they don't.

Re: Knights at g5 h5 by lc0

Posted: Sat Nov 10, 2018 3:36 pm
by Javier Ros
Werewolf wrote:
Sat Nov 10, 2018 11:53 am
I have long held a theory in my own personal over-the-board human chess which I haven't seen in a any chess book: that there is an advantage of having two knights side by side, especially if near the opponent's king.

Knights often make each other redundant, but when they are side by side, they don't.
In Megabase 2018 there are 4044 games with knights at g5 h5 with the following stats:

in 2343 games White won
606 games draw
in 1095 games Black won
White performance 2206
Black performance 2013

there are 2416 games with knights at g4 h4 with the following stats:

in 796 games White won
436 games draw
in 1184 games Black won
White performance 2113
Black performance 2209

The statistics corroborate your theory

Re: Knights at g5 h5 by lc0

Posted: Sun Nov 11, 2018 12:37 am
by ernest
Javier Ros wrote:
Sat Nov 10, 2018 3:36 pm
...
The statistics corroborate your theory
Very interesting !

But now I wonder, and why didn't I think of this before ? --->

In the famed 10 released games AlphaZero vs. SF8, a few remarkable moves by AlphaZero were found and analysed.

Does Lc0 now find any of these moves ?

Re: Knights at g5 h5 by lc0

Posted: Sun Nov 11, 2018 9:51 am
by Javier Ros
ernest wrote:
Sun Nov 11, 2018 12:37 am
Javier Ros wrote:
Sat Nov 10, 2018 3:36 pm
...
The statistics corroborate your theory
Very interesting !

But now I wonder, and why didn't I think of this before ? --->

In the famed 10 released games AlphaZero vs. SF8, a few remarkable moves by AlphaZero were found and analysed.

Does Lc0 now find any of these moves ?

Well, Alpha0 had an impact and attention in the media that lc0 has not had and I guess that few people have analyzed lc0 games.

For example the famous thorn pawns have appeared in lots of lc0 games

https://www.youtube.com/watch?v=Njzq0sR229c

https://www.youtube.com/watch?v=TxPpP_Tz2lA

and Alphazero, see 42.h6 in the following game



Another example are the holes at d6 f6 occupied by Queen and Rook, see the above game





and







And some positions that have reminded me a lot of AlphaZero style, of the type enclosed pieces, blocking and zugzwang: