The Knightmate Challenge!

[PK]

thanks for the alpha of winboard_f and for the Joker KM, I will share it with a couple of friends.

I was thinking a bit about knightmate-specific issues, and here are my two cents:

1. correct use of a commoner is all about square control. CCCP standard already maintains the table of squares controlled by pawns and uses it in quiescence pruning decisions - so I suppose that its knightmate strength would go up a bit if I implemented the same for the commoners.

2. commoner may be vieved almost like a general in shogi - it's invaluable defender of the king position. In many games Joker seems to get to the same conclusion by search, but perhaps it would be worthwile to implement it in eval too, as some kind of royal knight - own commoner proximity bonus.

3. I suppose that in the knightmate first move advantage means less than in chess, since it's doffocult to create a real pressure against the opponent's centre.

4. all programs, even Joker, do play extremely funny moves sometimes. it seems that in knightmate the move sequences are far less forced, so programs have more chances to steer towards ugly, unnatural game.

[hgm]

I am not sure this has anything to do with Knightmate: Joker80 also plays the occasional very funny move in Gothic Chess (such as moving its King for the purpose of destroying its castling rights and recieve the King-safety bonus for the Pawn shield in the center).

Joker is still a quite unbalanced engine, where I worked mostly on the search, with an eval that is mostly quick patchwork, even inferior to that of TSCP (e.g. no piece-square tables). Fairy-Max hardly has any eval at all, resulting in total disregard for doubled Pawns and passers. Funny enough, its loose attitude towards King safety seems to work quite well in Knightmate.

Using the Commoner to defend the Pawn shield might be a nice idea. I do wonder if a static defense is optimal, though, as it does not fully exploit the strongest asset of the Royal Knight, speed. I could imagine that you could build a strategy not based on hiding away the Royal Knight in a stronghold, but quickly shuttling it between 2 or 3 comparatively light shelters.

[hgm]

OK, one more tourney, and then I call it quits for Knightmate. (Unless there is a new engine, then I will add it to the tourney.)

The tourney also serves as a test for the automatic time handicapping: to make it more interesting, I have given CCCP-Knightmate and Fairy-Max 12-fold time-odds, and Joker KM 144-fold. All w.r.t. Dabbaba. Nominal time will be 40/12'. So against Dabbaba, CCCP and Fairy-Max will have 40/1', and Joker even 40/5". Against CCCP and Fairy-Max, Joker will have 40/1' (against their 40/12').

There will be 10 games per pairing, in a round robin.

May the best win!

[Jim Ablett]

OK, one more tourney, and then I call it quits for Knightmate. (Unless there is a new engine, then I will add it to the tourney.)

The tourney also serves as a test for the automatic time handicapping: to make it more interesting, I have given CCCP-Knightmate and Fairy-Max 12-fold time-odds, and Joker KM 144-fold. All w.r.t. Dabbaba. Nominal time will be 40/12'. So against Dabbaba, CCCP and Fairy-Max will have 40/1', and Joker even 40/5". Against CCCP and Fairy-Max, Joker will have 40/1' (against their 40/12').

There will be 10 games per pairing, in a round robin.

May the best win!

MsKcp 1.4 by Marcel Van Kervinck.

Modified to play 'Knightmate' chess variant
---------------------------------------
MsKcp also still plays normal chess.
---------------------------------------

What's New:
------------

MsKcp 1.4
----------
1. Added capability to play 'Knightmate' chess variant
2. Added time control.

http://www.zshare.net/download/12081499ad70755a/
Mirror:
http://mihd.net/xpihkro

Now wouldn't if be nice Winboard-F had another little window just for logos

Jim.

[hgm]

OK, I will add MSKCP to the tournament later, with the same time-odds handicap as CCCP and Fairy-Max.

As for the logos: I really don't see how they could be useful in WinBoard. This seems more a thing for the tournament manager and engine manager, where you could click on them to select the engine. WinBoard performs neither function.

Having a command-line argument /firstLogo="logos\engine1.gif" just to display a logo in a seperate window seems hardly useful.

[hgm]

On closer inspection (of the WBEC and RWBC ratings for MSCP) I put it in the tournament without any handicap. So far that seems to work out well, as the current standing against the 12-fold handicapped Fairy-Max is 1.5-1.5.

[Jim Ablett]

Mscp 1.41 update

1. Increased time safety margin. (noticed a few time losses)
2. Relaxed Royal Knight safety &
pawnshield in endgame.

http://www.zshare.net/download/12122954675ed5f0/


Jim.

[hgm]

Well, I let the old version complete the tournament, since it is not doing so bad. It did appreciably better against the 12-fold time-handicapped Fairy-Max as Dabbaba (it even managed to beat it). Now it is playing Dabbaba itself. MSKCP was leading 2-0, but then Dabbaba managed to win a difficult endgame, which, after the last Pawn disappeared, came to NbbNk. This is the Knightmate equivalent of KbbKn in normal Chess, and, just like the latter, it is a won end-game.

In fact, NbbNk is even mostly won with enforcement of the 50-move rule: logest DTC is 55 moves, but the fraction of positions that need more than 50 moves is totally negligible (0.0015%). This quite unlike KbbKn, where the bulk of the wins does take more than 50 moves (and under FIDE rules thus are draws).

[hgm]

The final standings of the time-odds Knightmate tourney are:

Code: Select all

                              Joker MSKCP CCCP  Fairy Dabba
 1. Joker KM / 144            ##### 00111 10111 =1110 01100
                              ##### ==101 10==1 10111 01110   64%  25.5 (745.0, 487.5)

 2. MSKCP                     11000 ##### 01001 =1001 1101=
                              ==010 ##### 11001 11001 01111   55%  22.0 (780.0, 402.0)

 3. CCCP new / 12             01000 10110 ##### 01110 1101=
                              01==0 00110 ##### 01010 11111   54%  21.5 (785.0, 380.0)

 4. Fairy-Max 4.8v / 12       =0001 =0110 10001 ##### =1011
                              01000 00110 10101 ##### =1111   50%  20.0 (800.0, 358.3)

 5. Dabbaba                   10011 0010= 0010= =0100 #####
                              10001 10000 00000 =0000 #####   27%  11.0 (890.0, 254.8)

Note that Joker KM was able to win the tourney despite giving a factor 144 time odds , while CCCP and Fairy-Max were playing at 12-fold reduced time. This should correspond to Elo reductions of ~500 and ~250, respectively.

So rough Elo guesses(using 1% = 7 Elo points) based on this tourney are:

Code: Select all

2398 Joker KM
2078 CCCP - Knightmate
2050 Fairy-Max
1835 MSKCP
1639 Dabbaba

[Jim Ablett]

Thanks for the tournament. I enjoyed it very much. I was up till 3am last night still watching.

Jim.