Author: Tony Werten
Date: 06:53:27 10/25/04
Go up one level in this thread
On October 25, 2004 at 08:21:52, Uri Blass wrote: >On October 25, 2004 at 04:22:36, Tony Werten wrote: > >>On October 25, 2004 at 03:27:00, Ed Schröder wrote: >> >>>On October 25, 2004 at 02:36:44, Tony Werten wrote: >>> >>>>Hi Ed, >>>> >>>>have you made a testrun of 1 game, (hope it reproduces) under the different >>>>settings. >>>> >>>>Most likely the logfiles will show a different (lower) knode speed. I can make >>>>up several excuses for that. >>>> >>>>- At startup, chessengine and interface are fixed to a different processor, >>>>meaning the second chessprogram gets the load of the interface as well (or any >>>>other process that the interface initialises). >>>> >>>>- Engine 2 is initialised on a crappy memory alignment. >>>> >>>>- Time related rounding errors are every time rounded in 1 the first engines >>>>favor. >>>> >>>>- All communication is handled as engine 1 first then engine 2 ( rather than >>>>engine on the move, engine not to move ) so during the whole game, engine 1 gets >>>>every information earlier than engine 2 >>>> >>>> >>>>If you feel the need, I'm pretty sure I can come up with some more :) >>>> >>>>Tony >>> >>> >>>But my procedure has been always 100% the same except for the loaded engine at >>>program start. Also I always reboot before I start a match. The issues you >>>address were valid in match-1, match-2 and match-3, still there is this >>>difference of 4.7%. >>> >>>What happened to Xinix? I can not believe your ranking, has fatherhood troubled >>>your astuteness? Programming while feeding babies? >> >>:) >> >>Even with 6 years of experience, 1.5 month isn't enough for a complete rewrite, >>changing from Delphi to C++, from single to dual and from x88 to bitboards. > >Why do you take task that is too hard for you? Because there's no fun in only doing the things you can do. > >If I understand correctly you write a lot of code and then test and discover a >lot of bugs. > >I think the correct way to program is to write something small and test that you >do not have serious bugs. > >other ways mean that you probably need to spend more time on your program to >have something free of bugs unless you are a genius. Well, thank you. It seems to be bugfree now. > > >> >>I could have played with an old version, but that would have been rather >>pointless. > >I think that it is pointless to play with a new version that you know that it is >weaker. > >I understand using a version that you are not sure if it is better but not using >a version that you know that it is weaker or a version that is totally untested. I do. Always better than playing with a version you know you are not going to develop any further. > > > The 2nd weekend was already a lot better, with the engine playing >>something that resembled chess, but there were still a lot of timing issues and >>so on. >> >>Every now and then, the search would explode, (not returning) and it would >>continue playing after a timeout of 6 minutes, while only actually having >>reached 8 ply. It blew a good position against Nexus with 3 of those in a row :( >> >>All in all, I'm not unhappy with the result from 2nd weekend. It'll be better >>next time. > >Unless you decide to do another rewrite. > >Personally I hate rewriting and I prefer to simply replace code step after step >and test that there are no bugs in every step. > >I think that if I have an hard problem that means a lot of code to write then it >is better not to attack the problem directly but think how to divide it to >easier tasks. Somehow I doubt that mixing Pascal and C++ and X88 and bitboards into one program will make it a lot clearer. Furthermore, it's much easier to implement multiprocessing in an empty program. Tony > >Uri
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.