U of A Computer Lines of Action Tournament            April 14, 2000

              YL Mon Via Dum Goa Mar Pee a.o Ear Glo Van Loa  Wins  L  D
  1 YL        x   2   2   2   2   2   2   2   2   2   2   2    22   0
  2 Mona      0   x   1   2   2   2   2   2   2   2   2   2    19   3
  3 Viagra    0   1   x   1   0   2   2   2   2   2   2   2    16   6
  4 Dumbie    0   0   1   x   0   1   2   2   2   2   2   2    14   8
  5 Goaloa    0   0   2   2   x   1   1   1  *0   2   2   2    13   9
  6 Maradona  0   0   0   1   1   x   0=  2   2   2   2   2    12   9  1
  7 Peever    0   0   0   0   1   1=  x   1   2   2   2   2    11  10  1
  8 a.out     0   0   0   0   1   0   1   x   2   2   2   1     9  13
  9 Earth     0   0   0   0  *2   0   0   0   x   1   2   2     7  15
 10 Gloat     0   0   0   0   0   0   0   0   1   x   2   2     5  17
 11 Vanline   0   0   0   0   0   0   0   0   0   0   x   2     2  20
 12 Loast     0   0   0   0   0   0   0   1   0   0   0   x     1  21

  * both games lost by forfeit (Goaloa was compiled with a wrong flag)

    Program       Author          Country
    -------       ------          -------
    YL        Yngvi Bjornsson     Iceland
    Mona      Darse Billings      Canada
    Viagra    Theodore Tegos      Greece
    Dumbie    Calin-doru Anton    Romania
    Goaloa    Rob Boyd            Canada
    Maradona  Ernesto Novillo     Argentina
    Peever    Steve Manners       Canada
    a.out     Loren Andruko       Canada
    Earth     Markian Hlynka      Canada
    Gloat     Marc Audy           Canada
    Vanline   Vanessa Yaremchuk   Canada
    Loast     Pablo Figuero       Columbia

Some thoughts and observations on the LOA tournament.

Yngvi's program was the class of the field, just as the crosstable
would suggest.  YL was in serious trouble in the first game against
Mona (later analysis showed the position was lost), but otherwise it
was a walk in the park.  This is not too surprising given that YL was
always searching 10-ply or deeper, with a strong evaluation function
(comparable to Mona's).  In fact, YL must have been rather unlucky to
even get into that one bad position.

Surprisingly, Mona was out-searching every other program in the
tournament, despite crawling along at only 20000 nodes per second.
Mona was always able to finish 7-ply (9-ply later in the game), and
was typically searching 2-ply deeper than the other programs, using
far fewer nodes, due to good move ordering and Null Move search.

The static threat analysis often enables Mona to detect a mate 1-ply
in advance, and it will go for a sequence of mate threats long before
it can prove the actual win.  Of course, this is a gamble, and it
back-fired in a big way against YL, throwing away a huge position to
make a cheap threat, which quickly petered out just over the horizon.
This is the second time Mona has lost this way to YL, so perhaps it
isn't as good a gamble as I thought. 

The strongest program in the heuristic search 657 class was a toss-up
between Viagra and Goaloa.

Theo's Viagra played two long, tough games against Mona, and hung in
long enough to get to a volatile "coin toss" position, which turned
out to be a win for White in both games.

Rob compiled his tournament version of Goaloa at 5 am, after a night
of heavy partying, and accidentally included a bunch of dead code.
This caused an illegal move and a segmentation fault in the match
against Earth.  Those errors had never occurred before (or since),
but his opponent did not offer to let him re-play the games, using
the real program.  If Goaloa had won those two games, it would have
tied for first among the programs in the class, at 13-3.

Goaloa had Mona busted in one game.  Mona announced a loss in 10-ply
(9-ply plus the static threat analysis), and Goaloa played the correct
move.  Mona announced a loss in 8-ply, and Rob said "Actually, I have
you winning...".  He had only gotten to 6-ply, so instead of finding
the mate in 7, Goaloa played a move that lost in 5, which Mona found
instantly.  A lucky escape!

Dumbie was "weird and wonderful", but quite effective, thanks to a
decent search.  All of its losses were after tough struggles.

Maradona played moves in one or two seconds, using a 5-ply fixed depth
search.  This may have hurt it later in the game, when an extra 2-ply
is often reachable.  Nevertheless, it did very well with what it had,
earning its solid result.

Peever played well early in the game (as did many of the programs using
a simple centrality evaluation), but it had trouble finishing, despite
completing 6-ply regularly.  The best example was a game it lost after
not seeing a winning connection in one move.  It also won the "Immoral
Move" distinction, for playing a help-mate in an overwhelming position. :)

a.out also got to 6-ply regularly, but the evaluation function could
have been stronger.  It didn't have much chance against the better
programs, but beat most of the weaker entries easily, on depth alone.

Earth suffered from an inefficient search.  In the two games against
Mona, it rarely got past 4-ply, even late in the game.  Like most other
programs, it threw away partial results from an unfinished iteration.

Gloat did surprisingly poorly, apparently due to bugs, rather than
inherent weakness.  In terms of strategic play, it was probably
stronger than some of the higher finishers.

Vanline had some technical glitches (ie. bugs).  It played okay as the
first player, but started making vacillating moves as White (eg. e2-g2,
g2-e2, etc).  It did, however, make legal moves and correctly detect
winning connections, which is more than some can say!

Loast was a side project, and wasn't fully developed at the time of the
tournament.  The intentional avoidance of moves that deliver mate may
have been a nasty little bug. :)

The gamescores for all of Mona's tournament games can be found here.