Author: Robert Hyatt
Date: 14:40:31 03/13/98
Go up one level in this thread
On March 13, 1998 at 17:32:45, Robert Hyatt wrote: >On March 13, 1998 at 13:52:45, Thorsten Czub wrote: > >>> Others aren't happy with no public autoplayer >>>from Fritz although no one said a thing about genius 4 which also had no >>>public autoplayer + a different version for SSDF testing. >> >>Please Bob, take notice: >> >>IF >>Genius 4 would have made an elo jump of 150 Elo points >>and IF >> I would have had it for a couple of months without recognizing >>this mega jump >>and IF it would have a non-public-autoplayer >> >>THAN i would have doubts of the same SIZE and loudness ! >> >>Please take car that it is NOT the fact that Fritz is suddenly no.1. >>By repeating this you simplify a much more complex problem. >> >>I have had fritz before Paris97. >>I had played games with it. >>I have now 64 MB hash and the junior-upgrade. >>And I don't get Enrique's results. >>I guess the reason is clear: he uses the "autoplayer" and I do play with >>my own hands. This seems to change the elo ! :-) >> >> >> > >note that just because Genius didn't jump to #1 doesn't mean that it >should >have been ignored. IE bad science is bad science. Either (a) there >should >have been a hue and cry when the genius4/dos version was tested to match >the hue and cry now, or (b) this is a moot point. > >But it makes no sense (at least to me) to only raise a ruckus when a >program >that is "different" does well, when the last program that was >"different" >didn't do so well... > >I go for consistency... nothing more... > >You are free to like what program you like, and to dislike what ever >programs you dislike... But I think it goes too far to accuse them of >wrongdoing with zero evidence. IE it would be nice for someone to take >one of the games Tony has published, and to go over it with the >commercial >fritz upgrade and prove that it would not play one or more of those >moves, >or that the timing or something would be greatly different, or that the >scores are different.. ie *something* to support the claims that have >been bouncing around... > > > >>>I think it is simply too emotional an issue. I've been on top of the >>>computer chess heap. I've been near the bottom. I've been everywhere >>>in between. The sun still rises tomorrow, and I still have just as much >>>fun today as I did when my program played it's very first move in >>>1968... >> >>All ok, but - as I said - you underestimate the problem. >>Fritz plays at my home like a slightly better Fritz4. >>Not more. No first rank, maybe 40 ElO's above Fritz4. >>But NO mega jump. > > >I haven't seen a "mega-jump" by *anyone*, to be honest. But I have seen >some new "speculation" in newer programs that just might not be the way >to >play against a deeper/faster searcher like Fritz... I also meant to add that I have been running an interesting experiment for the past 2 weeks. I have been running *identical* copies of Crafty, using *identical* books and everything (including hash size) with the only difference being the hardware. Crafty is still on the P6/200 that it always. Scrappy is on a PII/300, which reports itself to be 1.41 times faster than the P6/200. After about 700 games played by each on ICC, Crafty has been hanging in at 2700-2750 on ICC. Scrappy has been consistently at 2900-3000. I'm not quite sure what to make of this, because that 41% faster doesn't translate into that big a change. But, yet, I have gotten comments from computer operators (IE Max running ZarkovX) that "scrappy is a real problem for us where we were playing even or winning slightly more than 50%". I've gotten the same response from folks that are running other programs (ie genius 5 on a PII/333 for example.) So this fritz situation could be caused by lots of things, including the SSDF just *happening* to test it on the machine it runs the very best on. I'll report back with more data, but both crafty and scrappy have been playing the same "pool" from Roman to Dlugy to Shirov to the usual programs like ZarkovX, WchessX, Genius, Fritz and others. So *something* is going on here. Obviously 3,000 (3,022 at last check) is a rediculous rating for a computer, although that is a blitz rating. But the PII/300 and P6/200 are not *that* far apart. It's amazing that 41% is doing this. Or else a really long run of good/bad luck...
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.