Author: Robert Hyatt
Date: 11:38:19 07/01/99
Go up one level in this thread
On July 01, 1999 at 11:58:04, Peter Kappler wrote:
>On July 01, 1999 at 09:06:33, Robert Hyatt wrote:
>
>>On July 01, 1999 at 04:46:04, blass uri wrote:
>>
>>>
>>>On July 01, 1999 at 04:31:13, Stefan Meyer-Kahlen wrote:
>>>
>>><snipped>
>>>>>"most" of us don't make last-minute changes... except for opening book
>>>>>stuff. Which means that the version from 6 weeks prior to Paderborn is probably
>>>>>_very_ close to the version played at Paderborn, usually...
>>>>
>>>>Right now there is only one Shredder acount on ICC which is run by myself
>>>>("Shredder") and where you can play versions other than the commercial ones.
>>>>
>>>>Stefan
>>>
>>>Rober hyatt:
>>>>>"most" of us don't make last-minute changes... except for opening book
>>>
>>>Do the words "most" of us include you?
>>>
>>>Uri
>>
>>
>>For the two WMCCC events, absolutely. I generally 'freeze' things 90 days
>>before such an event. Allow minor tuning (eval changes) up to about 30 days
>>before the event, then 'live' with that thru the end of the tournament. In the
>>days of Cray Blitz I couldn't do that because we didn't get access to the cray
>>often enough and we often were 'tuning' right up to the tournament, which hurt
>>us often. More games are lost by last minute changes than all other cauases
>>lumped together. We used to joke at the ACM events that you could tell the
>>'beginners' because they were always logged on between rounds making serious
>>changes to their programs...
>
>
>Wasn't Crafty's sub-par result at Paris '97 the result of some last minute
>tweaks to the eval? Correct me if I'm wrong, but you said that the Paris
>version even lost a long match to GnuChess. Hard to believe that a well-tested
>version of Crafty would ever do that...
>
>--Peter
No... it was the result of long-term eval tweaking that paid much more
attention to humans playing anti-computer attacks than anything else. The
code was frozen well in advance... and at the time it was at an all-time high
on ICC. But it was simply mis-tuned. And for some reason, there were few
computers playing me during that 2-3 months so I simply didn't notice what was
going on. Once I saw it playing badly, I ran some tests here and proved that
it was most definitely playing poorly, as it would do _anything_ to prevent its
king from getting even the tiniest bit exposed.
The current version was treated the same way... Frozen months ago in
anticipation of going to Paderborn should the rules be 'bent'. But it didn't
happen.
I would _never_ show up at a tournament (now) with a version that had not played
several _thousand_ games on ICC with zero changes... which eliminates the silly
errors (screwed up underpromotions, screwed up time allocation, etc.) And I
then go over longer games carefully looking at output to make sure nothing
seems 'odd' in the program's behavior...
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.