This afternoon about 4 pm eastern standard time I'm planning to have Dragon play four games against my son IM Raymond Kaufman ("Mimsychess") on chess.com at knight odds at the same time limit as the Hikaru match, to test everything out. With that experience and then the Hikaru match we'll have a better idea of what is reasonable. I see no reason we can't propose a bullet match with Penguin after this; I don't know him but I can send him a message.
Dragon by Komodo Chess
Moderator: Ras
-
lkaufman
- Posts: 6303
- Joined: Sun Jan 10, 2010 6:15 am
- Location: Maryland USA
- Full name: Larry Kaufman
Re: Dragon by Komodo Chess
Komodo rules!
-
Mars
- Posts: 5
- Joined: Sat Nov 14, 2020 12:11 pm
- Full name: Martin Luzian
Re: Dragon by Komodo Chess
He seems to be a nice young guy, saw him recently on chess.com playing many bullet games against Farujia from Iran. Probably easier to convince him to play such a match first.
Although the following enounter over 117 games between him and Hikaru is already some time ago
I believe the latter is still 200-300 bullet points stronger than Tang (penguin).
-
Nordlandia
- Posts: 2845
- Joined: Fri Sep 25, 2015 9:38 pm
- Location: Sortland, Norway
Re: Dragon by Komodo Chess
Slightly deviate from the current discussion. Just asking whether dynamic contemt can be made adjustable as an parameter. Default contemt is static and both effect each other both ways.
-
Nordlandia
- Posts: 2845
- Joined: Fri Sep 25, 2015 9:38 pm
- Location: Sortland, Norway
-
Cornfed
- Posts: 511
- Joined: Sun Apr 26, 2020 11:40 pm
- Full name: Brian D. Smith
Re: Dragon by Komodo Chess
Question about the NNUE training. Does Komodo....err, Dragon, train on standard (not simple to calculate) endings?
If I could post a diagram I would (just don't know how to) but basically this...and asking because I won an ending like this today.
White (me) is a pawn up in a rook and pawn ending.
I foresee a way to get an outside passer on the queenside (f file).
White/Blacks kingside pawns are mirrors on h3/g2/f2 vs h6j/g7/f7
I know the stronger side will not want to let Black get in h5, so I play g4 and a few moves later the queenside pawns clarify to a c3 pawn and my rook on c5 with his on b3.
Question, would it pay to train 'standard winning attempts' via the NNUE....seems like there are a number of positions in early/mid-endgames where a search would find those positions and steer the engine toward the win which might be hard to find by mere calculation.
If I could post a diagram I would (just don't know how to) but basically this...and asking because I won an ending like this today.
White (me) is a pawn up in a rook and pawn ending.
I foresee a way to get an outside passer on the queenside (f file).
White/Blacks kingside pawns are mirrors on h3/g2/f2 vs h6j/g7/f7
I know the stronger side will not want to let Black get in h5, so I play g4 and a few moves later the queenside pawns clarify to a c3 pawn and my rook on c5 with his on b3.
Question, would it pay to train 'standard winning attempts' via the NNUE....seems like there are a number of positions in early/mid-endgames where a search would find those positions and steer the engine toward the win which might be hard to find by mere calculation.
-
lkaufman
- Posts: 6303
- Joined: Sun Jan 10, 2010 6:15 am
- Location: Maryland USA
- Full name: Larry Kaufman
Re: Dragon by Komodo Chess
Well, it trains on games. I suppose one could have Komodo (or Dragon) play out thousands of games from specified endgame positions and then train on those games, but coming up with a suitable collection of endgame positions is a lot of work unless one already exists (?).Cornfed wrote: ↑Sun Nov 22, 2020 11:14 pm Question about the NNUE training. Does Komodo....err, Dragon, train on standard (not simple to calculate) endings?
If I could post a diagram I would (just don't know how to) but basically this...and asking because I won an ending like this today.
White (me) is a pawn up in a rook and pawn ending.
I foresee a way to get an outside passer on the queenside (f file).
White/Blacks kingside pawns are mirrors on h3/g2/f2 vs h6j/g7/f7
I know the stronger side will not want to let Black get in h5, so I play g4 and a few moves later the queenside pawns clarify to a c3 pawn and my rook on c5 with his on b3.
Question, would it pay to train 'standard winning attempts' via the NNUE....seems like there are a number of positions in early/mid-endgames where a search would find those positions and steer the engine toward the win which might be hard to find by mere calculation.
Komodo rules!
-
JohnW
- Posts: 413
- Joined: Thu Nov 22, 2012 12:20 am
- Location: New Hampshire
Re: Dragon by Komodo Chess
Just wanted to say that I am really impressed how well Dragon has been playing against other engines I have installed on my PC. I really feel like I got my moneys worth in this purchase. Great job!
-
bastiball
- Posts: 5417
- Joined: Tue Oct 20, 2020 4:18 am
- Full name: Basti Dangca
Re: Dragon by Komodo Chess
How much the contempt affect Dragon? or it doesn't affect at all?
Basti Dangca
CCRL testing group
CCRL testing group
-
cc2150dx
- Posts: 469
- Joined: Sat Nov 30, 2013 9:51 am
- Full name: Jason Coombs
Re: Dragon by Komodo Chess
From Dragon readme file:
"A positive Contempt is used by the computer’s evaluation to discourage (or encourage if negative) piece trades, blocked pawn structures, and repetition draws early in the game."
"The default value 8 should help against weaker opponents. Against near-equal opponents e.g. (Stockfish 12/dev or Lc0) set Contempt to 0 for best results."
P.S. The second quote has been slightly modified.
-
bastiball
- Posts: 5417
- Joined: Tue Oct 20, 2020 4:18 am
- Full name: Basti Dangca
Re: Dragon by Komodo Chess
Thanks for the info, appreciate itcc2150dx wrote: ↑Mon Nov 23, 2020 2:42 pmFrom Dragon readme file:
"A positive Contempt is used by the computer’s evaluation to discourage (or encourage if negative) piece trades, blocked pawn structures, and repetition draws early in the game."
"The default value 8 should help against weaker opponents. Against near-equal opponents e.g. (Stockfish 12/dev or Lc0) set Contempt to 0 for best results."
P.S. The second quote has been slightly modified.
Basti Dangca
CCRL testing group
CCRL testing group
