Author: John Merlino
Date: 15:30:19 04/19/01
Go up one level in this thread
On April 19, 2001 at 17:38:42, Scott Woods wrote: >Hi John > >Hey - I just realised - I write to you more than I do my mother!! > I didn't need to know that.... ;-) > >I got 2 comps, Fritzmark ( only benchmark I got -sorry ) 323 and 359 - so a >slight difference. > > >I started testing std CM against Fritz 6 about a 9 days ago.. > So far 5.5 - 3.5 to CM8K ( both 634 Mg hash and over 128Mg memory ) > However about 40 games have been played and Fritz seems to time out > mid game with alarming frequency - > so....I have had to eliminate all suspect games and carefully > check that all games would be valid before posting... sod's law! > >Then about 5-6 days ago came posts of.... > >"NEW improved.. SSDF CM8K / DeepFritz" >(Which by the way I think something is still far wrong. No engine loses by that >margin ) And we think so as well, but we're not going to bother Hans until our testing confirms our suspicions. Additionally, the more games he plays, the more data we have to test with. > >and > >"Amazing win of Cm Ubtzinger over Deep Floggletoggle 4" >"Outstanding win of Cm Ubtinger....." >"Devastating win...." >"Stupendous win..." >"A cornecopia of wins..." >"Unbeatable..it still wins..." > > >Now, hands up - I am a bit green when it comes to computer chess. >But even I find such posts more than a bit sceptical. > As well you should, but Kurt's testing is persuasive. > >However, as I suspected that the standard switch settings on CM8K were not the >best I took the bait played them against each other. > >Not suprisingly....much of a muchness. > >However when playing Kurt's setting against another program it is a different >story. >I think that Kurt might have something here.. >(Although I won't commit myself to the extent of his enthusiam ;-) ) > I have never actually bothered testing a personality before, and have limited myself to book testing. In this case, though, I might make an exception, as we appear to have a true candidate for settings that are better than the default. HOWEVER, much more testing needs to be done at MANY different time controls. The CM default settings are tried and true, and we will be very wary of changing them without LOTS of testing. > > >If the Mattel, Mindscape, UBisoft? policy allows to you buy - Let's say Fritz 6. > >Look at the Fritz and Cm8K evaluations once the end of the opening book is >reached. With standard Cm8K the consensus of both engines is that CM8K is down >almost every game. With Ubzinger it is ahead almost every game. > Once again, some statistics would be useful here. This doesn't take very long to test, but I woulnd't be satisfied until I saw results from at least 1000 openings. > >If you wish I can play 100 opening book lines with CM8K & Fritz and CM8K >Ubzinger & Fritz .. ( Opening book lines are pretty quick to do. ) > >Wait - hell no - Kurt - They are your settings - you should drive this. >However - if you want a hand - let me & other CM nuts know (we are out there) > Yeah. Let Kurt run the show for a while (therefore he can be both author AND salesman!). I am in the process of further testing his (Kostick's) book. Then somebody needs to "independently" test his settings, then somebody needs to test his settings at lots of different time controls. Then.... > >I think the point is comparing how 1 set of parameters fairs against another >does not really mirror on how it compares against other engines. > > One thing I find fascinating about computer chess is watching > the "Thinking Lines" type windows. > CM can be plotting something along the Kingside whilst Fritz is > expecting a Queenside assault and vice versa. > >This is the real test. I'm not sure if it is a real test. I think the test is in the results, rather than the evals in the middle of a game. There are many more factors of a chess engine's ability to play than just its eval scores. However, it IS definitely fun to watch! I've played a few games as a guest on ICC against others who are manually playing computer programs, and swapping evals can be very interesting (and educational). jm
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.