Author: Christophe Theron
Date: 12:12:25 01/27/98
Go up one level in this thread
Hi Ed!
Tiger plays:
16. Rad1
(...Qe6 17. Nd6 Qxd6 18. Qxd6 Bxd6 19. Rxd6 b5 20. Bb3 Ke7...
time=54.92s, ply=11, score=+1.60)
>Strange thing is that the Sargon score till ply 8 was below -1.00
>Then on ply 9 the score went up to -0.43
Yes, strange. Tiger's score from ply depth 6 to 11 was always between
+1.42 and +1.60.
This remembers me the fact that Sargon's PV lines sometimes seemed very
strange, with a good move at the root, but expecting blunders from its
opponent, and with an obviously inaccurate score. Later, when I
developped my own chess program, I discovered that it happens when you
get a fail low or a fail high. And so I realized that Sargon's algorithm
only goal was to find the best move. As soon as it had demonstrated that
one move was the best, it didn't even care to know its exact score or
expected opponent reply. But not always. Sometimes, after having found
the best move and searched/rejected every other possibility, it went
back on the best to find out its exact score and PV.
Maybe it is what happened with 15...Nc6. On the first iterations, others
moves had scores clearly below -1.00, so Sargon has not spent much
effort on the exact value of Nc6. Maybe...
I also remember seeing strange phenomenas like: "Oh, this move fails
high, so it is the best. Let's have a look at the other moves. Oh
(again) this new one is even better! Let's have a look again at the
previous better...". So in the same iteration you could see Sargon
moving back and forth from one move to another, trying to determine
which is really the best, but avoiding to compute by how much it was
better.
I don't know any modern chess program using this technique any more. I
suppose all the hacks we do by using alpha and beta in forward pruning
would seriously confuse this algorithm.
The game is now:
[Event "Correspondance game"]
[Site "Internet"]
[Date "1998.01.01"]
[Round "?"]
[White "Chess Tiger K5 100Mhz 256Kb hash 1h/game"]
[Black "Sargon III Apple 2e 24h+/move"]
[Result "*"]
1. e4 c5 2. Nc3 b6 3. Nf3 Bb7 4. d4 cxd4 5. Nxd4 e5 6. Nf5 Nf6
7. Bg5 h6 8. Bxf6 Qxf6 9. Nb5 Kd8 10. Nfd6 Bc6 11. Bc4 Bxb5
12. Nxb5 Bb4+ 13. c3 Bc5 14. 0-0 a6 15. Qd5 Nc6 16. Rad1 *
Christophe
This page took 0.01 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.