Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Hiarcs7.32: "I am not Impressed"

Author: will smtih

Date: 21:12:10 09/22/99

Go up one level in this thread


On September 22, 1999 at 13:19:29, Ratko V Tomic wrote:

>There does seem to be a flaky or random aspect to H732.
>As you noticed, it takes queenside castle more often than
>one sees in other programs or GM games. It also makes
>occasionally adventuresome moves which are hard, if not
>impossible to reproduce later. I had H732 only for a month
>(and had H6 for couple years), but had decided to reinstall
>it few times and turn off any learning (after winning few
>games following some flaky dubious idea H732 tried and which
>was impossible to reproduce in replays).
>
>I suspect this flakeness is due to either the learning
>or permanent hash tables which perhaps don't get cleared
>(or initialized) consistently. It may be also that there
>is an intentional randmizer in the program, which benefits
>it most of the time, but occasionally causes a flaky move
>to be made. Since I am using it on 2 Pentium machines (one
>is 266 Mhz PII/MMX, the other 400 Mhz PII-Celeron, both with
>128Mb RAM, 64Mb set for hash tables, one Win95 another Win98),
>I doubt it is a machine or system software problem.
>
>But I have seen Fritz 5.32 do some flaky things as well, such
>as repeatedly playing and losing in one opening line (until
>I turned off the book learning and reloaded its book from the CD).
>
>Overall, though, I still find H732 most fun to play against
>(out of the CB engines; Rebel 10 is also fun to play against).
>Yes, it loses ocasionally due to some stupid irreproducable
>adventure it picks, but other programs make more positional
>blunders, especially in closed positions where e.g. Fritz 5.32 is
>at a complete loss what to do. I have also seen H732 avoid some
>greedy mistakes of Fritz (which goes after pawns while the attack
>is about to open up on its king side). Hiarcs somehow sees that
>pursuing a pawn would be bad, for no specific (tactical) reason
>within its search horizon. Only 20+ plies later, Fritz agrees
>that it has a problem.
>
>Even though my rating was only around 2100 USCF (when I played in
>competitions briefly as a graduate student, over a decade ago
>at Brown University), I find it with the current top programs that
>in most games I build a positional advantage for a while and then
>lose on a "cheap" tactical shot (which normally I should be able to see,
>but that would require consistent level of alertness over many moves
>and many games). Programs are like some disfigured athletes with
>hypertrophied tactical muscle and childlike strategic muscle. If they
>get you with their "strong arm" they win, otherwise your game looks
>better. Quite a difference from playing against a strong human master
>(e.g. my younger brother) or a grandmaster (I played occasionally
>GMs in friendly games; also in college had a neighbour [Damjanovic]
>who was an IM at the time and a youth champion of Yugoslavia and he,
>my brother, my sister and I had spent untold hours playing blitz, often
>till 3AM), where you're outplayed decisively in every aspect of the
>game and you can see clearly that you have no chance at all.
>Among the programs, Hiarcs (6 & &7) and Rebel (8-10b) are the closest
>to this humanlike balance, although they're still far from the real
>human player with similar formal rating.


What's your opinion of CM6000? does it play like a human?



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.