On Feb 23, 11:02 am, "
wrote:

I tried again to play a game to start to triangulate on a rating for
GitClub. Here's what happened. I thought I would try ChessMaster 10
with its rated personalities (who knows how good the ratings are and
under what conditions they were taken, but what the heck). I used a
personality with an 1193 rating and set the game for 2/6 Fischer time,
with CM as black, GC as white; GC at "easy" level (40 seconds per move
in theory). Here is the game:

White -- Black
(easy) -- (CM 1193)

1. d2-d4{34} e7-e5{24}
2. d4-e5{28} Bf8-c5{18}
3. Nb1-c3{26} Ng8-f6{20}
4. Ng1-f3{74} Rh8-g8{14}
5. e5-f6{20} Qd8-f6{12}
6. Nc3-d5{32} Qf6-d8{14}
7. Bc1-f4{32} d7-d6{28}
8. Qd1-d3{36} c7-c6{12}
9. Qd3-h7{42} Ke8-f8{12}
10. Nd5-c3{22} Qd8-e7{18}
11. Qh7-g8{20}

I never actually saw GC's 11th move; the board disappeared and GC
claimed a win when it had in fact just thrown away its queen (I looked
up the move in GC's move records afterwards). I guess GC thought this
was checkmate even though the white queen is unprotected.

Now, up to this point I was reasonably impressed. ChessMaster was
doing its usual stupid thing of throwing away material to lower its
rating, but GitClub was handily taking advantage of it (though when CM
threw away a knight GC waited an extra move to pick it up; luckily CM
left it hanging). Still, under these conditions thus far in the game
it seemed like GC was on the way to victory.

But with stupid bugs like claiming a checkmate in error, who has the
patience to try to evaluate GC properly, with a long enough series of
games under controlled conditions? I was about to do that but I was
frustrated again.

No doubt Sanny will fix this bug and proclaim the program better than
ever and ready to trounce Rybka in the future, but I'm about ready to
give it all up. After so many months the program ought to at least
play by the rules. And it would be really nice to show the final
position and not just delete the board and say "game over."


You know, it's funny how people suddenly
"show up" to report their test results when
the GetClub program is freaking out. Most
of the time, the program is playing a decent
game of chess and nobody "reports" on
their latest results, but when the serious
bugs pop up, invariably, several posters will
appear, complaining that this is the worst
chess program they've ever seen!

In my recent games, I have never seen the
program fail to make a one-move capture like
this, and in fact I had one position where the
program made a clever sacrifice, which was
to regain its material by force (I declined)
two moves later! Where it really has trouble
is in the endgame, and in misjudging spite
checks as something of real value. I think
the fact that Sanny keeps making reckless
changes is distorting the reporting here, by
encouraging reports by those who beat the
program like a carrot, while allowing those
who lose to remain silent; look at the Web
site itself and you will see that plenty of
players are losing games, but who are they
and where are their reports? Not here, that's
for sure! Not one player has come here to
"report" that they were outplayed-- ain't that
odd?


As for the program which hangs pieces as
quickly as possible to simulate the strength
of an 1100 player, that is a very poor way to
go about it. Human players tend to overlook
forks and mating attacks, but they don't just
move their pieces where they can be taken
by an enemy pawn-- except in scholastic
chess, way down below 1100. The 4. ...Rg8
move reminds me of little kids, who don't
quite know how all the pieces move yet;
these days, they might have ratings well
below 1000 at that level. And yet each time
its Queen was attacked, the program saved
it-- like someone who could play decent
chess. Who knows-- maybe this bizarre
combination of tactical alertness with regard
to the Queen, and suicidal tendencies with
regard to Knights balances out, resulting in
a net 1100 strength.

After Sanny said he had fixed the bugs
which were letting other people beat GC like
carrots (I was stuck in a won position where
the program refused to move, remember?),
I played another game-- a Benoni. Everything
seemed "normal" again; I sort of won a pawn
(don't count 'em; just look at how my pawn is
usable while its is superfluous) but ran into
some trouble with regard to my King's safety.
The program choked its opportunities, and
next thing I know I am mopping up a won
endgame, as usual. I would say it put up a
decent fight, but that its play was a bit
peculiar; no way is this thing anywhere near
USCF 1100 *strength*, although its style
could be compared to that of such a player.
More like 1700 or so is my guess. I think I
would far rather play a human 1500 player
than this monstrosity in a tournament; this
critter /can be/ dangerous, while the 1500
will not likely "see" anything I would miss.


-- help bot