Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: CCT3 is over

Author: Frank Phillips

Date: 11:40:48 05/30/01

Go up one level in this thread


I was hoping Johnhip  an ex work colleague who tested my early versions and was
watching while playing on Sunday (Monday for you?) would comment how weak the
first (many) versions were, although they got progressively better.  I suppose
as computers get faster and faster first versions will get better and better
against humans.

Getting rid of bugs is one reasons that puts me off starting again – in the
initial stages I think I used to do a re-write every six months or so and it
seemed to help development, increasingly cutting and pasting larger chunks of
the older version.  But the bugs tend to get more sophisticated and harder to
track down, although with post midnight programming after a day at work I still
get simple ones from typos.  Here is one just before CCT3
if (abs(best_w-best_b == 1)) that caused mayhem.   Part of code I added to make
it better and it got dramatically worse.  At least the probable culprit was
obvious.

Writing a validation routine (to compare data updated on the fly to that
regenerated from scratch) and a simple fixed depth search to compare nodes when
changes should not affect the shape of the search tree helped identify many
problems, particularly enpassant.  As does running under two different operating
systems.  The number of times I have tried to convince myself that different
nodes under Linux and Windows was somehow normal, is embarrassing.

What still surprises me is just how good the program was/is, even with major
errors – and whatever wacky evaluation term was stuck in last.  I can see the
logic in some things not being computable in practice and therefore evaluation
being important.  But I sometimes also think that computer speed is important to
_overcome_ parts of my static evaluation.



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.