On Sep 3, 12:48 am, Bjoern wrote:
Sanny wrote:
Fritz and Rybka I have not tested. Once my Program starts beating good
players like nomorechess, yourself, bob & Zebediah. Then I will test
it against Fritz & Rybka.

Once Bobby Fischer has beaten me in a match,
THEN I will allow Rybka to offer me pawn odds, but
not before. ;D


You should it would provide you lots of material showing where the
program can improve.

Now it do not make silly moves. Now you told it is not playing
intelligently. What do you mean by that. Can you show some moves which
you think were wrong.

Actually, looking at the programs past record, you should easily find
plenty of those, what I can access on the front page of your website
seems to indicated a record of
6 losses for master level
3 losses for normal level
4 losses for easy level
14 losses, potentially 1 draw (was that real or not?!) for beginner

and then there's the programs score of +5 -7 versus you personally,
Sanny, which appear to be the only games the program has actually won.

Why don't you just run all those losses through fritz?

I believe what Sanny has in mind is his Band-aid
approach whereby he finds an exact position early
in the game and then types in a specific move for
it to play by-rote. This is wrongheaded, and even
with a thousand such Band-aids the program will
be easy to beat, once beyond the Band-aid wall of
protection.

Personally, I don't know squat about Java, HTML,
or even writing a routine to generate a list of every
legal move in a given position, but it ought to be a
simple task to improve on the program's current
level, where it "resigns" a half-move before being
checkmated. Sheesh! Just do simple math:

K=99
Q=9
R=5
B=3
N=3
P=1

or this:

K=999
Q=100
R=56
B=34
N=31
P=10

and add them up for the "material score".


Then, after this is working smoothly, add on
a "positional score" which gives a bonus for
control of the center, King safety or whatever
you want, just make sure that you save the
original, tested-working version so you can go
back to Kansas if things don't pan out.

In my scenario, even a slow-as-molasses
Java module should be able to get well beyond
the current resign at mate-on-the-move state.
In my games, the program sits there for a while
and then, finally, realizes it is going to be mated
by force; the time for this ought to be essentially
*zero*, since there is only one legal move. For
instance, in the game where I recently beat the
master level I played ...QxR+, and it had just one
legal move, KxQ, after which I had ...R-h1 mate.

When I tried out another Web site which used
a Java chess program, there was no trouble and
it moved in good time, so I think the problem is
specific to Sanny's program/Web site.


-- help bot