The Knightmate Challenge!

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

Moderator: Ras

User avatar
hgm
Posts: 28468
Joined: Fri Mar 10, 2006 10:06 am
Location: Amsterdam
Full name: H G Muller

The Knightmate Challenge!

Post by hgm »

To celebrate the arrival of a new WinBoard engine that can play the variant Knightmate, namely Dabbaba, I have started a demonstration match in it between Dabbaba and Fairy-Max. Live viewing of the games is possible at my Battle-of-the-Goths webserver:

http://80.100.28.169/gothic/knightmate.html

Because Fairy-Max seems to be a bit stronger, I did include several time-handicapped versions of Fairy-Max in the gauntlet for Dabbaba, with time-odds factor 3 and 9 (next to the normal one). Nominal time control is 40/9', so the handicapped versions play 40/3' and 40/1'.

If anyone knows other WinBoard engines that can play Knightmate, please let me know, so I can include them in a tourney!
User avatar
Jim Ablett
Posts: 2423
Joined: Fri Jul 14, 2006 7:56 am
Location: London, England
Full name: Jim Ablett

Re: The Knightmate Challenge!

Post by Jim Ablett »

hgm wrote:To celebrate the arrival of a new WinBoard engine that can play the variant Knightmate, namely Dabbaba, I have started a demonstration match in it between Dabbaba and Fairy-Max. Live viewing of the games is possible at my Battle-of-the-Goths webserver:

http://80.100.28.169/gothic/knightmate.html

Because Fairy-Max seems to be a bit stronger, I did include several time-handicapped versions of Fairy-Max in the gauntlet for Dabbaba, with time-odds factor 3 and 9 (next to the normal one). Nominal time control is 40/9', so the handicapped versions play 40/3' and 40/1'.

If anyone knows other WinBoard engines that can play Knightmate, please let me know, so I can include them in a tourney!
Hi Harm,

Just got an email from Dabbaba author Jens B'k Nielsen.
He's watching the match :)

Jim.
PK
Posts: 913
Joined: Mon Jan 15, 2007 11:23 am
Location: Warsza

Re: The Knightmate Challenge!

Post by PK »

I have just created a knightmate version of CCCP. It comes with a minimal opening book of 3-6 ply variations to introduce some variety (I have not tested the openings).

Since it is TSCP-based, and I concentrated on evaluation, null move, qs prunings and hash tables, it still uses WB 1 protocol - hope it does not cause any problems under Winboard_F.

http://www.marittima.pl/cccp

regards,

Pawel Koziol
User avatar
Jim Ablett
Posts: 2423
Joined: Fri Jul 14, 2006 7:56 am
Location: London, England
Full name: Jim Ablett

Re: The Knightmate Challenge!

Post by Jim Ablett »

PK wrote:I have just created a knightmate version of CCCP. It comes with a minimal opening book of 3-6 ply variations to introduce some variety (I have not tested the openings).

Since it is TSCP-based, and I concentrated on evaluation, null move, qs prunings and hash tables, it still uses WB 1 protocol - hope it does not cause any problems under Winboard_F.

http://www.marittima.pl/cccp

regards,

Pawel Koziol

Great work Pawel! :)
Another engine to stomp on poor Dabbaba.
It'll be interesting to see how CCCP measures up against Fairy-Max. That does seem quite strong in Knightmate.
I was thinking there are a whole load of open source chess engines that could be converted to play Knightmate quite easily
and I think it's a great and fun variant. I hope it catches on.

regards,
Jim.
User avatar
hgm
Posts: 28468
Joined: Fri Mar 10, 2006 10:06 am
Location: Amsterdam
Full name: H G Muller

Re: The Knightmate Challenge!

Post by hgm »

I added CCCP - Knightmate to my tourney.

I start by having it play 20 games against Fairy-Max, then 20 games against Dabbaba. As I have no idea about the strength, initially all games will be without time odds (so 40/9'). Depending on the results, I might decide if time-handicapping would be needed, and for whom.

CCCP lost the first game against Fairy-Max, but one game of course means nothing.
User avatar
Jim Ablett
Posts: 2423
Joined: Fri Jul 14, 2006 7:56 am
Location: London, England
Full name: Jim Ablett

Re: The Knightmate Challenge!

Post by Jim Ablett »

hgm wrote:I added CCCP - Knightmate to my tourney.

I start by having it play 20 games against Fairy-Max, then 20 games against Dabbaba. As I have no idea about the strength, initially all games will be without time odds (so 40/9'). Depending on the results, I might decide if time-handicapping would be needed, and for whom.

CCCP lost the first game against Fairy-Max, but one game of course means nothing.
I just played some test games and Dabbaba seems to have the beating of CCCP also.
Are there anything special considerations when converting normal chess to Knightmate?
Does the eval have to be tuned a special way for Knightmate?
I'm not complaining, it gives Dabbaba an opponent it can beat, but CCCP should be stronger.
Has Pawel overlooked something in the conversion?

Jim.
User avatar
hgm
Posts: 28468
Joined: Fri Mar 10, 2006 10:06 am
Location: Amsterdam
Full name: H G Muller

Re: The Knightmate Challenge!

Post by hgm »

Fairy-Max is basically micro-Max that loads its move-generator table from a file, rather than having it as initialized data. (It also has some mode bits for each move to specify weir moves like hopping, but that is not used here, where all pieces are simple sliders or leapers.) Unlike Dabbaba, which really overwrites the tables for the pieces that you replace (or swap), it does support 15 different piece types, though, so that the exo-pieces can be present next to the normal ones. (Or you could make them all exo-pieces, if you want.)

The only evaluaion concern is the piece values, which are also read from the file. Micro-Max hardly has any other evaluation to speak off.

It seems to me, the major fault in CCCP's eval is that it overrates the Commoners (= the Knight substitutes). It did win the second game against Fairy-Max, though, with two Commoners versus R+B. But it was also a Pawn ahead, and had connected passers, while Fairy-Max had an isolated doubled Pawn. I think it was really the connected passers that were decisive, not the Commoners.
PK
Posts: 913
Joined: Mon Jan 15, 2007 11:23 am
Location: Warsza

Re: The Knightmate Challenge!

Post by PK »

Yes, commoners do seem to be the issue. My current material values are:

pawn=100, bishop = 345, commoner = 425, rook = 520, queen = 975.

I just stuck to the old adage that material value of the king should be about 4 pawns - but possibly this isn't true if the royal piece is a knight. CCCP KNIGHTMATE has also a bishop pair bonus, so after 1.d4 d5 2 Bf4 c6 it doesn't play 3.Bxb8. The rest of eval is basically mobility, pawn shield and two-dimensional tables adding bonuses for relative position in respect to the royal piece - perhaps again those tables are better suited against the normal king.

BTW, what material values do You use?
User avatar
hgm
Posts: 28468
Joined: Fri Mar 10, 2006 10:06 am
Location: Amsterdam
Full name: H G Muller

Re: The Knightmate Challenge!

Post by hgm »

Fairy-Max uses P=0.8, (N=2.8), B=3.2 (no pair bonus), R=4.8, Q=9.2. For Knightmate I set M (Commoner) = 2.4.

Dabbabah has P=90, N=310, B=325, R=500, Q=900, and then M=200.

4 Pawns for a commoner is cr*p, even in normal Chess (with normal Kings). In the opening they are worth less than a Knight (about 265 hen N=300), in the end-game (measured from 2 Knights against 2 Commoners, with 8 Pawns each on a 10x8 board) they were worth 320.
PK
Posts: 913
Joined: Mon Jan 15, 2007 11:23 am
Location: Warsza

Re: The Knightmate Challenge!

Post by PK »

I brought down the value of a commoner, as the games were too painful to watch. please download the recent version 1.01.06a - available at www.marittima.pl/cccp under the same link as the old one.