Author: Mark Young
Date: 20:22:02 06/24/99
Go up one level in this thread
On June 24, 1999 at 22:41:57, James T. Walker wrote: >On June 24, 1999 at 22:28:26, Mark Young wrote: > >>On June 24, 1999 at 20:04:29, James T. Walker wrote: >> >>>Hello, >>>Another way to prove it retains the hash tables is to turn off the permanent >>>brain (Pondering). Then when playing a game, notice the depth of search and how >>>long it takes to finda a move. When you play a move you will notice that Hiarcs >>>does not start it's search a 1 ply but may start its search at 5/8 or something >>>like that with 00:00:00 time showing taken for search. This indicates it is >>>using information from a previous search stored in the hash tables. This is why >>>Hiarcs has an advantage in engine/engine testing. The Pondering is turned off >>>for both engines but since Hiarcs does not reset it's hash tables it has >>>information left over to start the next search. But notice it's opponents >>>(Fritz,Junior,Nimzo) have to start every search from ply 1 even when the >>>predicted move is made. >>>Jim Walker >> >>I have not seen this, but I have not looked for it in engine vs engine games to >>see if it does retain hash from the previous search. >> >>To be honest I don't see this giving Hiarcs7 such a big advantage over the other >>programs at longer time controls, not to the extent that I see from other >>results that have been posted. >************ >I think some people reported early results when Hiarcs was doing good. Only a >long time and large amount of games can probe much about program strength. Of >course Blitz test will tell a lot in a short time since I can run at least 100 >games a day. Did you see my Blitz results? I only ran 50 games in the >engine/engine test and the results were Hiarcs came out stronger than in the >auto232 test. As far as I can see it is the only explanation for Hiarcs beating >Fritz in engine/engine testing. >************* > >> >>In my testing so far it has lost the blitz match I played at 10 secs. a move. >>It did crush Fritz 5, but junior 5 is giving it all it can handle at 30 secs a >>move, and 90 secs a move in the head to head match that I am playing now. >************* >You have probabley been told this before but it bears repeating. Playing games >at "10 seconds/move" is dubious because different programs have different >interpretations of this. Some take great latitude with this type time control >since it is not "Law". There are no lost games on time for instance so why not >take longer if needed? By contrast in my 200 game match between Hiarcs/Fritz, >Fritz lost 6 games on time. >************* > Time was an average of 10 secs per move, the level was set at 10'/60 >> >>And a quick note to Uri, Unless Hiarcs 7.32 is weaker then Hiarcs 7.01 or just >>got "lucky" I don't see how from the games so far... why Hiarcs 7.01 crushed >>Junior 5 in the SSDF testing. > >************* >I have this question in the back of my mind all the time. I don't own Hiarcs 7 >so I can't test it. I would like to hear from someone who has both programs and >has actually tested them not just against each other but each against identical >competition. >*********** >Jim Walker >***********
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.