RE: [computer-go] Words from the wise

["David Fotland" <fotland@xxxxxxxxxxxxxxx>]

I think you have to do whatever keeps you motivated.  If you fully polish
each 
algorithm before moving on, you will never have a program that plays well.
My 
approach was to do a simple implementation of each item, them over time
improve
the ones that caused bad moves as needed.

I wouldn't have started with the full board search though.  Without a good
evaluation,
alpha-beta search won't play good moves.  I suggest you work on the
evaluation,
including local search for string stability and connections.  When it can
make good
moves with one ply search, then go back to alpha-beta.

David

> -----Original Message-----
> From: computer-go-bounces@xxxxxxxxxxxxxxxxx 
> [mailto:computer-go-bounces@xxxxxxxxxxxxxxxxx] On Behalf Of john
> Sent: Friday, March 11, 2005 8:08 AM
> To: computer-go@xxxxxxxxxxxxxxxxx
> Subject: [computer-go] Words from the wise
> 
> 
> I've been writing a go bot for 18 months or so on and off.
> 
> It's gone from C, re-written as C++ OO style, re-written as C 
> just before Christmas.  Current name is 'Dumbbot' on KGS.
> 
> I've implemented alpha-beta, killer moves, iterative 
> deepening, hash table and full time-handling (interrupting 
> the iterative deepening without throwing away partial 
> knowledge gained at that level).  I've started to look at 
> connections + chains = groups and planned to expand that to 
> groups + weak associations = frameworks as two levels to make 
> strategic decisions against.
> 
> My problem is that all of these features are only 95% solid 
> and I spend more time introducing new features than robustly 
> testing the stuff I've already done.  I don't have the sort 
> of design-build-release mentality at home that I insist on 
> from others at work ;-)
> 
> My long-term intention is to develop a program that can play 
> well enough (and reliably enough) to consider exploiting 
> commercially in some way rather than _just_ having fun.
> 
> The question for you guys (and especially those involved with 
> commercial
> programs) is this:  should I bite the bullet, stop having fun 
> for a bit and put all those processes and tools in place 
> properly, or can I get to a strong program through a 
> combination of reasonable development discipline and sparky ideas?
> 
> I think I know the answer already :-( but your views and 
> experience would be highly valued.
> 
> John (kestrel on KGS)
> 
> _______________________________________________
> computer-go mailing list
> computer-go@xxxxxxxxxxxxxxxxx 
> http://www.computer-go.org/mailman/listinfo/computer-go/
> 


_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/