Author: Uri Blass
Date: 03:27:38 01/09/03
Go up one level in this thread
On January 09, 2003 at 05:51:47, Richard Pijl wrote: > >>>>I can announce that movei is not going to participate. >>>> > >Which is entirely up to you of course > >>>>It has enough tournaments that it participates and I see no need to participate >>>>in another tournament(it is better if I spend more time on fixing it and not on >>>>looking at more games of it). > >As I have participated in two tournaments (both in Leiden) I can assure you that >playing in those tournaments while watching the games will give you: >- Time to really think about every move your engine plays, question it and think >of ways to improve >- Comments of opponents/bystanders about what the opponent/bystanders see in the >position. >- General hints and tips by opponents/bystanders > >During (!) the last tournament (Dutch-open) I solved two bugs after round 1, one >more after round 3, and a whole bunch of bugs after round 6 (which was a whole >week). In round 8 Baron gave away a draw against Tiger (!), and in round 11 it >drew against the King. I'm very sure I would not have accomplished that without >the things you're forced to look at. > >Solving bugs is still the best way of improving your engine ... > >>> >>>I hope you will reconsider participating. I have found that the intensity of >>>tournament plays brings more bugs to light than any other way. > >Agreed! > >>>Also, new ideas for improvements seem to follow from tournaments, perhaps >>>because the author is more focused on the games. >> >>Movei is playing in enough tournaments. > >Even if you get the PGN _with_ the PV line, you still do not get the amount of >information that you will get when watching the game being played. Like moves >considered, fail-low/high sequences, score development in iterations etc. >> >>I do not see the big advantage from playing in another tournament. >>It is better if I work about what I need to do. > >There aren't many tournaments like this, so I usually grab the chance of >participating when I have the chance. >The todo list will be long anyway no matter how much you work on your engine. > >> >>There are a lot of free engines that movei can play against them. >>I do not understand why the ICC tournament can help me more than games >>that I get from other sources when the number of the games that I get from other >>sources is clearly bigger. > >It's not quantity that counts, but quality. Sometimes you need only one game or >position to find a lot of bugs. It's just a matter of careful watching what the >engine does. There is more to it than just the move it plays! > >> >>Games did not help me to find that movei use numextensions[-1] in the first >>iteration >>checkingbounds helped me to detect it. > >You should use all methods at your disposal to find bugs. Watch games being >played is just one method. Bounds checking is another method. Also compiling >your code with all warnings turned on can turn up bugs (type conversions to name >just one). >> >>For some reason when I use test postions I find that the number of nodes is the >>same even after correcting the error. > >Then you used the wrong testpositions for testing the improvement :-) > >>There was also another error(not deterministic error) before the error of >>numextensions[-1] but for some reason I cannot reproduce it now. > >Games will turn up a lot of those, and even more when you're start doing >cut-offs based on the transposition table. >Non-deterministic errors in test positions are IMO pointing to uninitialized >memory or out of bounds memory and most of those should be found with a bounds >checker. The problem is that the non deterministic error happened only when I used bound checking and not in normal run of the program. Here is the descreption of the error that cheking bound told me: Subscript Overrun: Array index value 44 exceeds array index limit of 8. The problem is that no array that I used at the same time had dmenesion of 8 and that the error now cannot be reproduced. > >>I believe that discussions here can help me more than games to get rid of these >>errors. > >You will have those discussions online during tournaments ... > >>The main problem of movei except it is not bugs but lack of knowledge. >>I looked in a lot of previous games of movei and I do not believe that new >>tournaments are going to help me. > >If you have bugs in search or eval, correcting those will get you more than >adding more knowledge. The main jumps in strength of the Baron are caused by >solving bugs, not by adding new knowledge ... (adding knowledge usually >introduces new bugs :-) ) > >Richard. It depends on the bugs I try to be careful about it and for every code that I add to add code to check for bugs. The baron has clearly more knowledge than movei(that until today have no knowledge about king safety or pawn structure) and I plan to add knowledge about pawn structure but I will be careful to give bonus only in clear cases The first step that I plan to do is to give bonus for passed pawns only if it is protected by a pawn but I need first to update my bitboard for pawns when I make and unmake moves and check that I do it correctly and only later when I have the bitboard for pawns correctly I may continue and define the bitboard for protected passed pawns. 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.