DIR Return Create A Forum - Home
---------------------------------------------------------
The Chess Variant Forum
HTML https://chessvariantforum.createaforum.com
---------------------------------------------------------
*****************************************************
DIR Return to: Correspondence Game Requests
*****************************************************
#Post#: 315--------------------------------------------------
Re: Gothic Chess Correspondence Game
By: chilipepper Date: January 31, 2018, 10:09 am
---------------------------------------------------------
[quote author=GothicChessInventor link=topic=53.msg310#msg310
date=1517390670]
So don't put too much faith in programs...
[/quote]
I still think competitions between engines is pretty cool. Even
if a human can beat a software program, a program that can win
over all other programs is a pretty cool invention! 8)
#Post#: 316--------------------------------------------------
Re: Gothic Chess Correspondence Game
By: Greg Strong Date: January 31, 2018, 10:48 am
---------------------------------------------------------
[quote author=GothicChessInventor link=topic=53.msg310#msg310
date=1517390670]
Chess V is very easy to crush. I gave it 3 minutes per move
tonight on my Intel i7-5960X overclocked to 4.6 GHz and I was
able to beat it by giving it tons of material in positions it
obviously did not understand.[/quote]
Thanks, Ed. When I get a little time I'll go over this and see
what I can learn ...
#Post#: 317--------------------------------------------------
Re: Gothic Chess Correspondence Game
By: Greg Strong Date: January 31, 2018, 11:11 am
---------------------------------------------------------
[quote author=HGMuller link=topic=53.msg312#msg312
date=1517407655]
I am not sure what exactly Gregory means by 'beats it
regularly'. Does that imply more than half of the time?[/quote]
Yes, sorry, that's what I meant. I was surprised to be able to
reach this point since Fairy-Max sees about three times as many
nodes per second and ChessV has a pretty basic evaluation also.
The difference must be primarily in the search algorithm.
Chessv has most of the advanced stuff: PVS, IID, TT, Null move,
LMR, razoring, futility, and delta pruning. It is missing
ProbCut and singular extension - not sure how much those are
worth. And, of course, it's not multi-threaded, although I do
plan to do that. The Stockfish 'Lazy SMP' approach looks pretty
easy.
Although, chilipepper, if you are looking for strong 10x8
engines, also check out Joker80, another of H.G.'s engines.
Although I haven't tested it recently, it's probably still
stronger than ChessV. To use it with chessv is easy. Just make
a new "Joker80" sub-folder under chessv's Engines\XBoard folder,
and throw the joker80.exe in it (download here
HTML http://home.hccnet.nl/h.g.muller/joker80.exe.)
Next time you
fire up chessv, it will notice the new engine, fire it up,
interrogate it about what variants and features it supports, and
will make it available as an option for any supported games. Or
if you prefer to store the exe somewhere else, you can use
Engines > New > Discover Engine.
#Post#: 318--------------------------------------------------
Re: Gothic Chess Correspondence Game
By: HGMuller Date: January 31, 2018, 11:45 am
---------------------------------------------------------
Ah, so you made good progress. Fairy-Max of course has stayed
the same forall these years (in the variants it already
supported), so it is a good opponent to guage that. BTW,
Fairy-Max also has IID, TT, null move, LMR, delta pruning. I
guess the most limiting factor for it is poor move ordering
(hash move, and then the rest of the move in move-genration
order).
#Post#: 321--------------------------------------------------
Re: Gothic Chess Correspondence Game
By: ebinola Date: January 31, 2018, 1:40 pm
---------------------------------------------------------
[quote author=GothicChessInventor link=topic=53.msg304#msg304
date=1517363517]
[quote author=ebinola link=topic=53.msg300#msg300
date=1517352569]
I'll take a crack at it. Not to say that I'll win but I'd like
to play nonetheless.
[/quote]
OK I'll play 1. d4 if you want to play against me. It was not
fair for me to volunteer Panzer :)
[/quote]
Are you wanting to do it in this thread?
If so, 1... h6
#Post#: 325--------------------------------------------------
Re: Gothic Chess Correspondence Game
By: GothicChessInventor Date: January 31, 2018, 3:57 pm
---------------------------------------------------------
[quote author=HGMuller link=topic=53.msg312#msg312
date=1517407655]
Note that Fairy-Max is what I consider a weak program. It is
derived from micro-Max by just changing the possible board sizes
and piece moves, and micro-Max was optimized for being small
(maximum Elo per character) rather than for being strong. As a
result virtually all Chess knowledge was culled out of it, and
the search algorithms are also pretty basic.
[/quote]
I would like to add here that I believe MicroMax may hold the
"world record" for fewest characters for a fully-functioning
chess program forever. I have tried my own minimal character
count just to try and get close to what H.G. achieved, and I
couldn't get better than 2.2 times his code length. This is a
remarkable achievement of H.G.'s and it should be recognized for
what it is!
[quote author=HGMuller link=topic=53.msg312#msg312
date=1517407655]
Furthermore, there seems to be a strategic aspect to 10x8 Chess
that is not very important in orthodox Chess, and makes it
relatively easy for Humas that know the trick to beat most
programs. So using the same algoritm as an orthodox engine, just
enlarging the board size, does not automatically preserve the
strength.
[/quote]
This is 100% true and 200% true if you are talking about Gothic
Chess. In part of my game design process, I wanted players to be
"torn" between having to make one of two irreversible move
choices that would influence the play into the "forever future."
Examples:
1. d4 d5 2. Nh3 g6
Now what for white? If 3. Nc3 then 3...Nc6 strikes twice at the
d4 pawn, so do you play 3. c3 here? Let's see:
3. c3 Nh6 4. g3 Nc6 5. Bg2 looks fine.
Let's try Nc3 instead.
3. Nc3 Nc6 4. g3 and now white tasks black with the same dilema:
what to do about the d-pawn under double attack?
4...Nxd4 5. e3! Nc6 and white has a few ways to recapture.
The idea being, players can choose a line of play with strategic
aims right from the opening, and this influences the entire
nature of the game. It is BEYOND the scope of what computers can
determine, even if they could complete a 30-ply search, it would
barely scratch the surface of white lies beneath.
#Post#: 326--------------------------------------------------
Re: Gothic Chess Correspondence Game
By: GothicChessInventor Date: January 31, 2018, 4:01 pm
---------------------------------------------------------
1. d4 h6
2. h3
Sure this thread is fine. Thanks for accepting. Good luck.
#Post#: 327--------------------------------------------------
Re: Gothic Chess Correspondence Game
By: HGMuller Date: January 31, 2018, 4:18 pm
---------------------------------------------------------
[quote author=GothicChessInventor link=topic=53.msg325#msg325
date=1517435874]I would like to add here that I believe MicroMax
may hold the "world record" for fewest characters for a
fully-functioning chess program forever. I have tried my own
minimal character count just to try and get close to what H.G.
achieved, and I couldn't get better than 2.2 times his code
length. This is a remarkable achievement of H.G.'s and it should
be recognized for what it is!
[/quote]
Tough luck, but the micro-Max record is already beaten: Oscar
Toledo G. wrote a Chess program that is only about 1280
characters, while the smallest micro-Max version was at about
1480, IIRC. I believe the program is called Toledo nano-Chess.
It is much weaker than micro-Max, however. I never was aiming to
be absolutely the smallest, because that will unavoidably lead
to something that is also asymptotically weak, barely better
than a random-move generator. And that is not very interesting.
So I always optimized Elo/char, allowing me to add extra
characters if they coded for a feature that would bring enough
Elo. Nevertheless I did hold the record for being smallest for
some time. And using some of Oscar's tricks could also shave
many characters of micro-Max.
In Fairy-Max I no longer care much about Elo/char, and many
characters were added just to expand the type of moves it can
support, without this bringing any Elo in variants that do not
involve such strange moves like hoppers or bent sliders. This
added some 30% extra code lines in the AI. But the evaluation
remained minimalistic, and it is a miracle it plays as well as
it does. Having no knowledge is often better than having wrong
knowledge, and I guess that is what saves the day. Many muh more
complex programs are weaker because some of the knowledge they
have is wrong.
#Post#: 333--------------------------------------------------
Re: Gothic Chess Correspondence Game
By: Greg Strong Date: January 31, 2018, 10:13 pm
---------------------------------------------------------
[quote author=HGMuller link=topic=53.msg318#msg318
date=1517420712]
I guess the most limiting factor for it is poor move ordering
(hash move, and then the rest of the move in move-genration
order).
[/quote]
Ouch. Yeah, that would hurt. I bet just trying captures before
non-captures would make a huge difference. Even if it meant
running the move generator twice, once for captures, and then
again for non-captures (if there is no cut-off), I bet that
would still be a win, and would keep your code fairly simple.
Or you could consider doing this only at likely cut-nodes...
#Post#: 334--------------------------------------------------
Re: Gothic Chess Correspondence Game
By: HGMuller Date: February 1, 2018, 2:53 am
---------------------------------------------------------
Well, in a sense it does search captures before non-captures,
because of the IID, which starts at QS depth, where all
non-captures are skipped. So obvious material gains (i.e.
visible in QS) will be preferred as cut-moves over non-captures.
And before the QS iteration the IID even does an MVV/LVA
iteration, to select the best capture to start the QS iteration
with. For each iteration it runs the move generator again, as
there is no storage of moves. It only remembers which move is
best (so the next iteration can start with it).
*****************************************************
DIR Next Page