Author: Robert Hyatt
Date: 07:03:03 02/02/04
Go up one level in this thread
On February 01, 2004 at 23:34:29, Peter Kasinski wrote: >On February 01, 2004 at 22:39:20, Robert Hyatt wrote: > >>The main preparation items were the hardware (supplied by AMD in Austin, TX) a >>quad 848 box with 8 gigs of RAM. I sent one of my 15K U320 SCSI drives out >>there with the 3-4-5 piece tables and the opening book stuff. >> >>The opening book prep was pretty minimal. Gower and I (and a couple of chess >>players he brought with him) tested a couple of QG-type lines to be sure that we >>had reasonable choices for black's responses to d4/c4 openings. As black we had >>elected to play the 2. ... e6 variation of the Sicilian that we used so many >>years with Cray Blitz. It gives a reasonable position with a bit of asymmetry >>to produce a position that isn't "dead" right out of book. >> >>For round 9, since a win was critical, I chose to vary and not play 1. d4 since >>Crafty has been known for that forever. I decided to go with 1. e4 and if black >>allowed it, I was going to play the Evan's gambit, an opening I like for >>computers as white. But Hiarcs didn't cooperate and we ended up in a Guico that >>transposed to a Ruy. >> >>The book has two components. The base book is my old stand-by enormous.pgn >>product made by "book create enormous.pgn 60 10". I then had a bookc.bin with a >>few (and I do mean a few as in 6-7 lines for white, 2-3 for black) lines to >>guide the thing into reasonable positions. I didn't try to win out of book, but >>I did try to avoid losing out of book. :) >> >>Now on to the rounds. >> >>Round 1. White vs messchess. Normal 1. d4 opening and we dropped out of book >>on move 15 with an evaluation of +.67, totally acceptable. :) The score jumped >>by about .3 every move and by move 40 the opponent resigned at +17.4. Good >>start. NPS was ridiculous of course. First search was to 17 plies deep. >>Seemed to hit 14+ the entire game. >> >>Round 2. Black vs Pharaon. A QGD (Lasker's) that popped out of book looking >>drawish as most Lasker's do. However starting around move 30, black used some >>endgame/pawn structure knowledge to produce a significant "edge" that won a pawn >>by move 35 and the usual score jump after every move resulted in a resignation >>at move 47. Typical depth here was 16-17 plies but queens were off. >> >>Round 3. White vs Pepito. Another 1. d4 opening and another +.7 out of book, >>but Crafty has always liked 1. d4 (+.7 is pure positional scores, not a pawn in >>material ahead). This was pretty stable through move 40, but slowly Crafty >>improved and around move 40 things starting to get significantly better. By >>move 44 we were at +2.5 and while the game went on for 30 more moves, it was >>over basically... >> >>Round 4. Black vs Zappa. First bad book line. We had accidentally left a >>response to d4 for black that led to a position we didn't like. Somehow I did >>not delete it when we were testing, or I deleted it locally and not on the >>Opteron. In any case, we had no real winning chances in this game, and Crafty >>found a forced perpetual that ended the game at move 30. >> >>Round 5. White vs Searcher. We had a gross network problem and I could not >>connect to ICC or the opteron, fortunately the opteron and ICC were not affected >>and the game started without me. Crafty simply played very well here in another >>1. d4 except this time we were about 0.0 out of book as black tried something >>different. The evaluation steadily climbed, but slowly, until move 20 saw the >>first fail-high winning a pawn plus. Things fell apart quickly and the game >>ended at move 32, at +8.20... >> >>End of day 1. Four wins, one draw. Some amazing search speeds. Some decent >>luck with the book to avoid "book losses". Not a bad day at all. >> >>Round 6. White vs Yace. Expected to be one of the strongest competitors as >>usual. Out of book on move 7 and Crafty was happy. So happy it offered the >>b-pawn as a gift, although Yace declined. By move 12 this was +2. By move 20, >>+3.5. This is the _wrong_ kind of position to play into against a machine as >>fast as the quad 848 box. Yace resigned at move 35 at +10.0. An easy win >>thanks to a bad book line choice by Yace, that turned what promised to be a real >>struggle into a quick tactical bust. It's happened to me before, so I know how >>it feels. :) >> >>Round 7. Black vs ChessThinker. Another QG by white but leading to a pretty >>good white position this time after we chose a bad reply. By move 25 this was >>+.75 with us as black not liking that very much. The score quickly reached +1 >>and stuck around there until around move 40 where it had reached +1.5 and >>eventually +2. But Crafty finally got a bit more consolidated and the score >>slowly started coming back down. I thought we were going to lose this game when >> the score hit +2 around move 45. By move 55 it was down to .75 and continued >>to drop until the game ended in a hard-fought draw. >> >>Round 8. White vs Junior. Obviously this would be a bear of an opponent, and >>the game did not disappoint. Crafty worked up an edge after leaving the book at >>+.17 (nearly equal) to +.6 by move 16 (book ended at move 7 as Junior chose to >>not repeat the previously tried responses to d4 others played, and apparently >>listened to a comment by IM Schroer to play something different. Junior let >>things get a bit wild, and starting around move 20 we were hitting 16 plies >>every time and our score was up to +1.0 with some wild tactics. We ended up >>winning a piece for three pawns and then a very sharp tactical struggle followed >>where either side might have won. A well-deserved by both players result of >>draw was reached (perpetual by Junior) at move 64. >> >>Round 9. White vs Hiarcs. Another difficult opponent. Before starting this >>round, I chose to do something different with the book for two reasons. First, >>our 1. d4 had become pretty predictable and I didn't want to repeat the Junior >>game and possibly draw again.. I was concerned that had Ruffian won, a draw by >>Crafty would have given the title to Ruffian and I chose to try something a bit >>more dangerous and go with 1. e4. In fact, this was the game where I had set up >>for the Evan's Gambit but Hiarcs didn't play 3. Bc5 but went to the two knights. >> Had I prepared a bit more, I might have gone for something wild there as well, >>but I had not prepared any other 1. e4 openings. The Guico line ended at move >>15 with a perfectly balanced position, both programs showing almost exactly 0.0 >>for the evaluation. By move 25 the evaluation was up to +.75 as Crafty is not >>prone to "sit" on a position having been brought up in the GM quagmire of ICC. >> It continued to press this positional advantage into winning a pawn, but it >>reached a difficult to win (if it was winnable) ending with an extra pawn, with >>one rook left on the board. After a long struggle, this ended in a draw. >>However, I like games where Crafty is playing for the win and the opponent is >>struggling for the draw, rather than the opposite (which happened to us twice as >>black as previously mentioned.) >> >>Final result was 5 wins, 4 draws, 7.0/9.0 for the final result. Hiarcs and >>Crafty were tied, and Zappa put Ruffian down for the count to join the group >>with 7.0. >> >>The playoff was not something I particularly like, as the main event was 45 10 >>while the playoff was double RR 5 3 blitz event. I played Zappa first and Zappa >>played a couple of dubious book lines and Crafty won both games (not easily, but >>it had enough horsepower that you give it a pawn, you can expect problems if you >>don't have some significant compensation. In the other two blitz games Crafty >>played, it drew hiarcs in both. The first was a bit of a struggle as Crafty >>chose a dubious book line but it was fast enough to hold on anyway. The second >>was also a rather bad Sicilian line and again Crafty struggled for the draw, but >>a draw was all it needed to win the playoff. >> >>Final impressions were many. >> >>1. Competition is tough. Everyone is getting stronger. You can't take a >>single game for granted any longer, in general. >> >>2. The AMD box is simply amazing. 8M-11M nodes per second. Most games >>averaged 8M or so throughout, after both sides have castled, 7M or so before >>both sides castle. Search depths were typically 12-16 plies, depending on the >>position. Open kings and lots of checks dropped the depth to 11-12 a very few >>times, forcing lines in the Junior game let Crafty actually search as deeply >>(reported ply depth) as Junior even though Junior counts plies differently. >> >>3. Luck still plays a part. whether it be luck that avoids a bad book line for >>you, or produces a bad book line for your opponent, or just searching deep >>enough in a critical position to see a win or avoid a loss, etc. >> >>4. Debugging is critical. No point in losing games due to bugs, screwed up >>time management, etc. Lots of games on a chess server can weed those things out >>nicely. >> >>5. A book is important, but not as important as "some" claim. My normal big >>book did just fine, with about 100 moves total in my "start" book to select >>particular openings and avoid others. >> >>6. Perhaps that "if they thought they had a chance, they would have come" >>nonsense can now be put to rest for all time. I ran on a machine that might >>have been about 1/4 the speed of the machine I could have used in Graz. Anybody >>that thinks that would not have been "competitive" is out of their mind. >> >>I've been doing these competitions since 1976. They are _still_ fun. :) >> >> >>And for Vincent, let me add the following to avoid all the discussions he will >>start: >> >>1. My evaluation sucks. >>2. My parallel search sucks. >>3. My book sucks. >>4. My simple q-search sucks. >>5. My wife is ugly. >>6. I'm too old and stubborn to have a chance. >>7. My kids are stupid. >>8. My mother used to wear army shoes. >>9. My truck is 9 years old. >>10. I don't know squat about NUMA. >> >>I think that about covers it. >> >>:) >> >>Now he doesn't have to add his two cents' worth, assuming he has two cents' >>worth to add. :) > >Congrats and thanks - this was a lot of fun! > >As always, I am left wondering - perhaps you could comment on this: I am trying >to imagine what Crafty could do by now if _half_ of your professional >competition gave _you_ access to their source code... But, of course, this is >not entirely necessary to have fun, as you have repeatedly demonstrated. I really don't know. Back in "the good old days" we swapped code all the time. IE Ed Kozdrowicki (COKO), Fred Swartz (Chaos), Dave Slate (chess 4.x), Tom Truscott (Duchess) and so forth. I didn't spend a lot of time trying to study their code as the programs were simply too big. (Coko was over 100K lines of Fortran, for example). I learned much more by talking to them. On the phone. And often. There is less of that today. IE Bruce and I have had many phone conversations about various chess ideas. That is by far the best way to improve your own code. IE in my case, were I not the author of Crafty, I might read the source to see the comments, since it is at least 50% comments, but I would not try to understand the code as it would take a lot of time that could be better spent working on my own stuff. > >So, good luck and many happy returns to you and your clearly dysfunctional >family! > Thanks. :) >:-) > >PK
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.