Author: Robert Hyatt
Date: 07:17:04 10/14/99
Go up one level in this thread
On October 14, 1999 at 01:40:46, Ratko V Tomic wrote:
>> you ought to play a program with a 1 ply depth limit, which is basically
>> relying totally on that 'evaluation' you mentioned. I think you'll find
>> that the evaluation is far stronger than just a "1 ply search with no
>> evaluation at all".
>
>Well, in all these years (since Challenger 7, back in 1981) never played at 1
>ply level setting. But for the sake of test, and past my bed time, I played
>couple games at "1 ply level," one against Fritz 5.32 and one against Crafty
>16.6 (which was included in the Fritz shell, both programs used General.ctg book
>which came with Fritz). Fritz got checkmated in 16 moves, Crafty in 20 moves,
>the two little childish games are included below. I think, whatever evaluation,
>it can't be but simple-minded overall (granted, some special patterns may
>trigger extra calculations based on very precise knowledge, e.g. some not well
>known endgame pattern matching, if they can ever get there). Who would make
>those kinds of nps otherwise. On 400 Mhz Celeron, Fritz maybe spends at most few
>hundred asm instructions per evaluation to get the 400 knps. So at one ply
>setting, as you can see below, they fall for much too easy traps. (BTW, once out
>of the book, does your version of Crafty, at 1 ply level reproduce moves that
>the 16.6 makes?)
you are missing the point. The eval 'guides' the program into positions that
are favorable. But it assumees that the search has scouted out all the hidden
tactics and taken care of those. Sure you can beat the program tactically at
1 ply. But notice the moves they make. If there were no tactics, the moves
look very reasonable, doing the usual things like protecting pawn structure,
occupying the center, etc.
I have played many 'beginner' chess players on 1 ply searches and they don't
have a chance. Many run crafty on ICC and try to weaken it with sd=1, and they
+still+ can't get its rating below 1600-1800 at the very best. In fact, in
blitz, it is probably a 2200 player even at that shallow depth. How much time
did _you_ take per move? Or did you use your knowledge + your ability to look
ahead to beat the program with knowledge only?
Try it in very fast games, where you can't really out-calculate it easily,
to see how your knowledge stacks up against its knowledge. You might be
surprised. Or to stop them from overlooking instant mates, try 2 ply
searches...
>
>Since Fritz seems to be easier at this level, I wonder if anyone here would try
>beating it in fewer than 16 moves at "1 ply level."
>
>
>[Event "Level=1ply. "]
>[Site "Lexington, MA"]
>[Date "1999.10.14"]
>[Round "?"]
>[White "Tomic, Ratko"]
>[Black "Crafty 16.6"]
>[Result "1-0"]
>[ECO "C42"]
>[PlyCount "39"]
>
>{32768kB, GENERAL.ctg. C400
>} 1. e4 {0} 1... e5 {0} 2. Nf3 {2} 2... Nf6 {0} 3.
>Nc3 {5} 3... Bb4 {0} 4. Bc4 {15} 4... Bxc3 {0} 5. dxc3 {12} 5... d6 {0} 6. Qd3
>{10} 6... O-O {0.21/1 0} 7. Bg5 {4} 7... Nc6 {0.34/1 0} 8. h3 {8} 8... a6 {
>-0.19/1 1} 9. a3 {4} 9... Bd7 {-0.35/1 0} 10. O-O {10} 10... Qe7 {0.04/1 0} 11.
>Nh4 {14} 11... Rad8 {-0.01/1 0} 12. f4 {5} 12... b5 {-0.32/1 0} 13. Ba2 {6}
>13... exf4 {-0.41/1 0} 14. Rxf4 {6} 14... Ne5 {-0.55/1 0} 15. Qf1 {20} 15...
>Bc6 {-0.05/1 0} 16. Rxf6 {14} 16... gxf6 {-0.54/1 0} 17. Bxf6 {3} 17... Qd7 {
>0.53/1 0} 18. Qf4 {9} 18... Rde8 {-0.51/1 0} 19. Qh6 {8} 19... Bxe4 {-1.58/1 0}
>20. Qg7# {3} 1-0
>
>[Event "Level=1ply. "]
>[Site "Lexington, MA"]
>[Date "1999.10.14"]
>[Round "?"]
>[White "Tomic, Ratko"]
>[Black "Fritz 5.32"]
>[Result "1-0"]
>[ECO "B00"]
>[PlyCount "31"]
>
>{65536kB, GENERAL.ctg. C400
>} 1. e4 {0} 1... Nc6 {0} 2. d4 {2} 2... e5 {0} 3.
>d5 {4} 3... Nce7 {0} 4. d6 {2} 4... Nc6 {0.31/1 1} 5. dxc7 {7} 5... Qxc7 {
>0.06/1 0} 6. Nc3 {3} 6... Nf6 {0.06/1 1} 7. Bg5 {7} 7... Bb4 {0.06/1 1} 8. Qd3
>{9} 8... Bxc3+ {-0.16/1 1} 9. bxc3 {3} 9... Ng4 {-0.09/1 0} 10. Nf3 {11} 10...
>O-O {-0.16/1 1} 11. h3 {4} 11... Nh6 {0.31/1 0} 12. Bxh6 {4} 12... gxh6 {
>0.28/1 1} 13. Nh4 {7} 13... b6 {0.12/1 1} 14. Nf5 {3} 14... Bb7 {0.66/1 0} 15.
>Qg3+ {7} 15... Kh8 {#1/1 0} 16. Qg7# {2} 1-0
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.