Dragon MCTS in the Giuoco Piano (6.e5)

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

Moderator: Ras

KLc
Posts: 140
Joined: Wed Jun 03, 2020 6:46 am
Full name: Kurt Lanc

Dragon MCTS in the Giuoco Piano (6.e5)

Post by KLc »

Does anyone know why Dragon (2.5.1) in MCTS mode prefers 6.e5 (+0.15/18) to 6.exd4 (0.02/18) in the Giuoco Piano 1.e4 e5 2.Nf3 Nf6 3.Bc4 Bc5 4.c3 Nf6 5.d4 exd4 but "no" other engine does, e.g. Stockfish has a 0.00/46 for both 6.e5 and 6.exd4. The 6.e5 is also recommended in the Kaufman repertoire.
lkaufman
Posts: 6279
Joined: Sun Jan 10, 2010 6:15 am
Location: Maryland USA
Full name: Larry Kaufman

Re: Dragon MCTS in the Giuoco Piano (6.e5)

Post by lkaufman »

KLc wrote: Thu Jan 06, 2022 9:13 pm Does anyone know why Dragon (2.5.1) in MCTS mode prefers 6.e5 (+0.15/18) to 6.exd4 (0.02/18) in the Giuoco Piano 1.e4 e5 2.Nf3 Nf6 3.Bc4 Bc5 4.c3 Nf6 5.d4 exd4 but "no" other engine does, e.g. Stockfish has a 0.00/46 for both 6.e5 and 6.exd4. The 6.e5 is also recommended in the Kaufman repertoire.
Top players today clearly agree with Dragon here. The old move 6.exd4 is generally considered to lead to a draw by repetition (...Nc6-a5-c6) after a few moves, while 6.e5 is a serious try for advantage. Sure they both lead to a draw ultimately with perfect play, but 6.e5 presents more problems for Black.
Komodo rules!
KLc
Posts: 140
Joined: Wed Jun 03, 2020 6:46 am
Full name: Kurt Lanc

Re: Dragon MCTS in the Giuoco Piano (6.e5)

Post by KLc »

I see, thank you! So, is this actually just part of your evaluation function and nothing special to MCTS?
Odd Gunnar Malin
Posts: 310
Joined: Wed Mar 08, 2006 9:59 pm
Location: Norway, Vads?
Full name: Odd Gunnar Malin

Re: Dragon MCTS in the Giuoco Piano (6.e5)

Post by Odd Gunnar Malin »

Also two of the new repertoire here prefere 6.e5, Wesly So in his Chessable 1.e4 repertoire, and Gawain Jones in his Coffeehouse repertoire.
lkaufman
Posts: 6279
Joined: Sun Jan 10, 2010 6:15 am
Location: Maryland USA
Full name: Larry Kaufman

Re: Dragon MCTS in the Giuoco Piano (6.e5)

Post by lkaufman »

KLc wrote: Thu Jan 06, 2022 10:50 pm I see, thank you! So, is this actually just part of your evaluation function and nothing special to MCTS?
Well, it may very well be due to MCTS. The normal Dragon assumes perfect play by both sides (as well as it can judge it), and may see Black equalizing in one particular variation after 6.e5 (rightly or wrongly). MCTS does not assume that it can judge perfect play, it assumes some likelihood of seemingly slightly inferior moves getting chosen, so it is more likely to recommend a good practical choice even if it is not better against supposedly perfect play. Whether that applies to this particular decision would be very hard to determine, but it seems likely to me. You might say MCTS is more humble; A/B engines are more "arrogant"!
Komodo rules!
KLc
Posts: 140
Joined: Wed Jun 03, 2020 6:46 am
Full name: Kurt Lanc

Re: Dragon MCTS in the Giuoco Piano (6.e5)

Post by KLc »

Very nice. Thanks for explaining this!
Andrew
Posts: 231
Joined: Thu Mar 09, 2006 12:51 am
Location: Australia

Re: Dragon MCTS in the Giuoco Piano (6.e5)

Post by Andrew »

4 ... Nf6 is illegal, is this Nc6?

Thanks!

Andrew
KLc
Posts: 140
Joined: Wed Jun 03, 2020 6:46 am
Full name: Kurt Lanc

Re: Dragon MCTS in the Giuoco Piano (6.e5)

Post by KLc »

Ah, sorry, I meant 2...Nc6 (just main line Giuoco Piano). I can't edit this anymore.