Author: Robert Hyatt
Date: 15:16:55 07/10/01
Go up one level in this thread
On July 10, 2001 at 17:57:12, Peter Berger wrote: >I am not really familiar with computerchess events and experiments prior to 1998 >unfortunately as I only started to follow the development then . > >Youzr current Scrappy experiment is getting old as you pointed out yourself . > >Scrappy and Crafty are idle often enough that another experiment which would >interest me very much would be feasible . > >Set your formula to one neutral , allow-anyone-over-2000 one , like >Postmodernist's : > >formula = "autocolor & (unrated | (rating >= 2000)) & registered & !freeweek & >!timeodds & (time > 0) & ((!blitz) | time > (inc-5)) & ((wild=0) | (wild=5))" > >This is already slightly changed : PM allows 1800+ but I think this doesn't make >too much sense in general ( there _are_ some very able anti-comp players at that >very low rating range but there are lots more of time-wasting experiences at >this low end ) . > >Crafty is expected to do very well against the anti-comps low-rated masters like >the great "kako" et al . I am positive you've been there and done that but it >has been _that_ long ago ; so it has been forgotten or never seen by some >newbies. > >Seeing some of them playing against the computers has always been a good cure >for hyperbole and a valid entry to the "computers are GMs" discussion to my >experience . You have made big efforts as I have learned here to teach your >program how to do much better than others in this category but you rarely have >shown your program's ability recently . > >This would be a really interesting approach for the scrappy account ; much more >interesting than your "rating pig" ( Moreland expression ) "competition" with >"spitfire" . > I'm not in any "competition" with spitfire at all. ICC has had _plenty_ of "human-only_ programs. The bullet record was held for a long time by a program that would only play 2 0. Same for blitz. Scrappy was done for two reasons. One, I used to run it on very slow hardware, to see how the faster hardware changed the search and the results. The second was the "experiment" I started last year... to see how the same program, on the same hardware, would perform against computers and against humans. I had been specifically tuning for human-caused problems, and I was curious. It was pretty obvious that I _was_ doing much better against humans than against computers. And I decided to stop the experiment a while back. I wasn't expecting the messages and requests from the players that were regularly playing it however... they wanted it back as they could play it as often as they wanted, without computers breaking in or playing long games (ie Brazillianmaster has played 8-10 standard games against crafty in a row, tying it up for hours at a time.) That was the reason I put it back on... It had _already_ set the ridiculous high-rating mark for blitz and bullet, and I don't know that it will ever reach those heights again as several humans have caught on to anti-computer plans and can at least draw often enough that it will probably not reach 3550+ again... It isn't about ratings at all. It was about testing a hypothesis I had, as well as burning some unused computer cycles. Scrappy runs at "nice 20" for those that know what that means, which means if anybody else uses that machine. scrappy gets bombed... as far as the 5 3 formula goes, I chose that number a few years ago after asking GM players what they preferred to play. Most said 5 0 with some 5 3 times thrown in. I don't particularly like longer blitz games as they really are not blitz games. I try to track the rating as I make changes, such that a drop generally means a problem was introduced. But if it plays 5 0 and gets to 3300, and then plays nothing but 5 12 the next day and drops to 3200, then what caused the drop? Longer time controls or a bug? Too many variables to make this understandable. At 5 0 through 5 3, there is little difference. THe games are fast, the human gets 3 secs per move added so he need not flag, and the "blitz" games go pretty quickly. I hadn't thought about Bruce's point particularly, but it is valid. But by the same token, all the "tiger clones" and "crafty clones" and "fritz clones" and "shredder clones" and so forth are _also_ a problem to those of us developing chess engines. There are so many that they really dilute the chances to get to play GM players, because the human operators challenge the GMs all the time, keeping them occupied. I would _much_ prefer to see "authors-only" on the servers, for obvious reasons, perhaps with certain time periods for exceptions. But those would be limited... I can't see why a GM would play a 2800 Tiger when he can find a 3100 version and have a much better chance of bumping his rating. This has turned ICC into a giant rating war. But it has been so since I have been on there. We just used to have dozens of gnu's, and then dozens of Crafty's. And now dozens of _everything_. >I hope you don't get me wrong - this message is dead serious , appreciates your >achievements very much and is one _most_ serious begging for a possible and very >interesting event to follow. > >pete
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.