Author: Uri Blass
Date: 07:15:23 04/23/02
Go up one level in this thread
On April 23, 2002 at 04:43:00, David Dory wrote: >On April 23, 2002 at 02:11:34, Gian-Carlo Pascutto wrote: > >>On April 22, 2002 at 19:40:00, David Dory wrote: >> >>>The programmer for Hiarcs stated point number 2. Perhaps you could throw your >>>considerable logic at him? >> >>The closest I can find is this: >> >>The Hiarcs 7.32 search was prone to exploding in the middle game at depths >>beyond ten plies and this resulted in a search which did not take good advantage >>of today's machines. >> >>I can read into this as saying that the engine did not perform well when >>given a long time to think. Which amounts to saying that it did not play >>well at _longer_ timecontrols or when machines got _faster_. >> >>This is *directly* in contradiction to your own statement about Hiarcs >>just one post earlier. >> >>I don't really see the statement on the ChessBase as a factual conlcusion, >>since it's in the middle of the salestalk for the program and there isn't >>any data to back it up. >> >>>My point was that relatively faster engines will perform somewhat better than >>>relatively slower one's, >> >>I don't believe this is true. >> >>>>What you said implies that Hiarcs gets stronger at longer timecontrols, which >is haven't seen any evidence for . . . >>> >>>ALL programs get stronger with more time to think, what the heck are you saying? >>>Are you saying Hiarcs plays worse when given longer to think? >> >>Change 'stronger' into 'relatively stronger compared to other programs' in my >>sentence. >> >>>I don't have an interest in the palm, or follow those posts - so I have no idea >>>what clue you're referring to here. >> >> >>-- >>GCP > >I had not seen the post on CCC yet about Mark U.'s interview on Hiarcs8 when I >replied to your post. It was from Mark, but it was regarding Hiarcs overall >chess playing philosophy (especially for 7.32.)** , and it was reported in the >British computer chess mag. > >Yes, I still believe Hiarcs7.32 got stronger relative to other programs when >playing at time controls longer than blitz. Maybe your impression is based on comparing hiarcs's results on good hardware and on slow hardware. Hiarcs may earn more speed from faster hardware and on the same hardware that Fritz is 2 times faster Hiarcs may be 3 times faster(these are not exact numbers but only the idea). It can help hiarcs to be better in blitz on p800 and not on p450 but it does not mean that it is better on long time control. I saw some strange mistakes of hiarcs7.32 at long time control when it learned wrong information from previous search(I do not know what it does because I am not the programmer of hiarcs but it plays moves that the only way to reproduce them is to startplaying from a previous move that it did not blunder). It does not trust the wrong information that it learns at high iterations but unfortunately at long time control it happens when it has not enough time to find a better move when it usually has enough time to do it at blitz time control because the branching factor at blitz is better. I believe that this reason(without the higher branching factor) is enough to cause hiarcs to be weaker relative to other programs at long time control. There are also other reasons(for example the fact that it never extends to more than 31 plies at least based on the selective search that I see). It is not very important but I believe that it also can cost 10-20 elo at tournament time control when it costs nothing at blitz. 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.