Author: Peter McKenzie
Date: 22:09:17 05/28/01
Go up one level in this thread
On May 28, 2001 at 23:44:47, Dann Corbit wrote: >On May 28, 2001 at 19:45:52, Peter McKenzie wrote: > >>On May 28, 2001 at 16:18:06, Frank Phillips wrote: >> >>>On May 28, 2001 at 13:30:44, Bruce Moreland wrote: >>> >>>>On May 28, 2001 at 08:33:19, Peter Berger wrote: >>>> >>>>>I think the CCT3 was a great and enjoyable event ! The atmosphere was very >>>>>friendly and there were no debates this time . >>>>> >>>>>A few ideas and suggestions for future events though: >>>>> >>>>>a.) Every engine participating should be able to and do write logfiles . These >>>>>should be sent to some arbiter right after the game has been finished in case of >>>>>debates. >>>>> >>>>>As far as I know every participant would have been able to do so ( for example >>>>>for the WinBoard programs the winboard.debug would be sufficient ) . >>>> >>>>There are some weird people out there who either want to take credit for other >>>>people's work. I think there have been some others who don't realize that their >>>>own work is pretty insignificant compared to what was already present in what >>>>they started from. >>> >>> >>>This is interesting from the other end so to speak. I start with Marsland’s 'A >>>review of Game Tree Pruning' (said anatomy of chess program elsewhere), that >>>too, all the relevant internet basic articles about move generation, various >>>type of search, GNU chess source and tscp. After trying to figure out basic >>>alpha-beta, Negamax, Mtdf (heaven forbid), 0x88, mailbox, bitboards, piece >>>square tables, and many headaches, I (and really I mean one, as in >>>hypothetically) decide on mailbox, basically follow the reciepe in Marsland >>>guided by tscp and end up after some time and effort with a basic engine. Then >>>add hash tables, then killer moves, history moves etc etc. Later null move. >>>Following discussions on CCC add sophisticated tweaks like IID, adaptive null >>>moves. All this gets refined and improved as time goes on and understanding >>>increases. Parallel developments include move ordering using captures first >>>sorted by SEE, then over time (again guided by discussion on CCC and articles) >>>full blown ordering into hash move, winning and even captures, killer moves, >>>non-captures sorted by history and finally losing captures. And so on and so >>>on. Finally the whole entity is like Marsland described but with modern tweaks >>>and hence fundamentally like Crafty, Dark Thought, GNU, etc, but not exactly >>>copied as in at a point in time and the source looks entirely different. >>> >>>Similar story about evaluation. I hear about outside pawns, the lack of >>>knowledge of which has cost me dearly. Look at Crafty’s source. Talk on CCC. >>>Do not understand Crafty source but implement the well explained algorithm in my >>>own way. So it goes on. Again, program tends to Crafty but not exactly >>>copied. On the surface it looks entirely different. Maybe no bitboards at all >>>for example. Finally I start to implement independent ideas (tweaks really) and >>>it begins to diverge again. >>> >>>At which stage in this process, if ever, is one allowed to turn up to a >>>tournament and say here I am with my program, rather than my version of >>>Marsland/tscp/GNU/Crafty/Heinz/Kerrigan/….? At any time my contribution to the >>>fundamental approach is minor. I did not invented minimax, negamax, alpha beta, >>>Zorbrist hashing, bitboards, mailbox, 0x88 or whatever? Sure I use them. But I >>>might equally well, I suppose, have just taken Crafty (after all it is better) >>>and messed around with its code or just the data and weightings. Heck there is >>>even an input file for doing this now (I think?) so I do not even need a >>>compiler. >>> >>>I have no point here. Nor do I now know where I stand. Maybe however I now >>>understand what Will was really asking :-/ >>> >>>Frank >> >>Nice post Frank, and congratulatons on your result in CCT3. >>I think your approach is abolutely fine and sensible and legitimate. >> >>I do wonder how much time you saved being guided by TSCP and GNU Chess. I'd >>estimate that starting with say TSCP, even if you rewrite every line of code in >>it, would still give about a one year head start over starting from scratch >>without any source code as reference. >> >>Why? Because what you get: >> >>- a clean framework and structure >>- a complete working test bed >>- a relatively bug free implementation >>- a decent starting point for an evaluation function (sensible tunings etc) >> >>There is just so much you can get from source code that you don't get from >>straight descriptions or explanations. >> >>This is not a criticism in anyway, just me thinking out loud. Heck, if I was >>starting a chess program today then I'd probably start with TSCP or something >>similar. In fact, there is so much crap in LambChop now that I might even do a >>new program built up from something like TSCP! > >I'd start with SCP instead. It does 200K NPS right out of the gate, and is no >larger than TSCP. Well I was only half-serious, but what is SCP anyway?
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.