Author: Robert Hyatt
Date: 21:03:51 12/20/01
Go up one level in this thread
On December 20, 2001 at 18:53:38, Dann Corbit wrote: >On December 20, 2001 at 18:30:51, Ulrich Tuerke wrote: > >>On December 20, 2001 at 17:56:17, Tom Kerrigan wrote: >> >>>I can't think of a reason why commercial programmers would have an edge over >>>amateurs when coming up with good ideas/techniques. >> >>When an amateur programmer comes up with lots of good ideas/techniques, then >>he'll have success and go commercial eventually. I think that's how Shredder and >>Junior proceeded. > >And Tiger -- probably all the commercials started out like that. At least most >of them. > >But some amateur programs seem to prefer to stay that way. Yace and Crafty are >as good as some professional programs. If the authors of amateur programs got >to dedicate the time, many of them would become professional quality. > >I think probably the biggest difference is that huge blocks of time allow not >only thorough innovation but also great debugging effort to make solid code. > >I think a good example of the excellence of mistake free code is TSCP. It has >no hash table, opening book, no move list. Really none of the advanced >techniques are in there. And yet it is stronger than many programs that have >those features because it is robust. I would bet, that if I could take my testing methodology back to the 1970's, that my program would have won 90% of the ACM events. The thing that set the winners apart was _always_ "robustness". If you were repairing bugs between rounds, you were not a serious contender. If you only played a few games a year, then you also could not be a serious contender. Crafty is _extremely_ robust. It plays too many games to have serious bugs. Yes it does have several small ones that show up from time to time. But if you go back and look at acm events, you find things like pawns reaching the 8th rank and not promoting, castling bugs (like the new fritz), etc. I don't see those in (say) CCT events. For Cray Blitz, we were _severely_ limited in available test time for machines that cost 60 million dollars. And we paid for it many times. We had our share of good luck (1983 and 1986) and we had our share of bad luck too. All a direct result of insufficient testing. If I could have had ICC-type testing in the 1970's while others could not have, the chess history books would have looked _far_ different than they do today. There is nothing better than testing, other than _more_ testing.
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.