Sanny wrote:
Many years ago, a fellow came from out
of nowhere with achessprogram that was
stronger than all the others. It had a nice
GUI, nicerchesspieces than the number
two program, and most curious of all, it fit
on a single floppy disk as it was quite
small in addition to searching faster.

Even GetClub Chess Program will fit in a floppy Disk

GetClub Chess Program Size is Just 200 KB. While a Floppy can take 1.2
MB. So you can pack 5 Chess Programs in a floppy.

The thing is that the size is small but the complexity is very high.

I think GetClub program is good enough for all Human Players. Only the
player using other commercial programs are beating the higher levels.

That seems very unlikely given that it is so predictable about
sacrificing N for 2P and going for spite checks. Its endgame play is
complete rubbish (but then even the top engines are still a little bit
flakey there at tournament time controls) I'd be surprised if it could
beat most serious club players reliably looking at the recent games.

I think it will now comfortably beat all casual chess players, but I
think anyone at ELO 2000+ will crush it (excluding careless slip ups).

Only you are able to win the Master Level rest all are not going
higher than Normal Level.

That is more because the humans die of boredom waiting for GetClub to
make a move at the higher levels. You would do well to find a couple of
local club level chess players with some mathematics or programming
background if you actually want to make the program stronger.

Playing toy matches and tweaking things isn't the right approach.

If you are serious you need to get a set of test cases that it should be
able to solve and regression test the engine against them after each
"improvement" - that way you can be sure you *are* making things better.

Some of the positions I have already picked out would be good ones to
have in your test suite.

A first step towards this is to allow the engine to accept FEN notation
for game positions and set up the board at the test position. Then a
modification so that you can display a table of move score in cp ranked
by evaluation as a function of search depth.

And underneath the time taken, number of nodes processed and terminal
nodes evaluated (this may require adding diagnostics to your code).

eg.
Ply: 1 2 3 4 5 6
e4 05
d4 03
c4 00
Nf3 -01
....
Nodes
Evals
Time
(this table should be sorted by the deepest evaluation available)

The objective is to try and get the ply 1 evaluation to be self
consistent with the deeper evaluations and so minimise the number of
nodes the program has to evaluate to get to a given depth.

The maximum depth examined by selective extensions is also possibly
useful here too. We need to see how it is thinking and what its internal
evaluation is for the moves to be able to give sensible advice.

Regards,
Martin Brown
** Posted from http://www.teranews.com **