Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Thank you for the info, Peter. Appreciated. (NT)

Author: Uri Blass

Date: 13:52:35 09/11/00

Go up one level in this thread


On September 11, 2000 at 14:59:52, Mogens Larsen wrote:

>On September 11, 2000 at 14:20:33, Chessfun wrote:
>
>>>Here is one blunder of Gandalf4.32f against Crafty
>>>
>>>[D]r4rk1/3p1ppp/1pb1p3/p1b1P2B/2P2B1q/P1N5/1PQ2PPP/3R1RK1 w - - 0 1
>
>>Maybe Christian could check, but others in limited time modes would
>>alos play this move: As the time controls were 40/40' it can be guessed
>>the move took maybe 2 minutes.
>
>That is a reasonable assumption. Probably slightly less.
>
>>Crafty 17-13N takes 3:30 on my machine to switch to Bg3.
>>Fritz 6b takes 0:46 to switch to Bg3.
>
>Gandalf analysis:
>10440 <first :      2 94 0 254  g4 Qh3
>10440 <first :      2 97 10 515  b4 !!
>10440 <first :      2 257 10 606  b4 axb4 axb4 Qxf4 bxc5 Qxe5
>10490 <first :      3 203 10 2531  b4 axb4 axb4 Bxb4 Na2 Qxf4 Nxb4 Qxe5 Nxc6
>dxc6
>10550 <first :      4 60 10 10447  b4 axb4 axb4 Bxb4 Na2 Rxa2 Qxa2 Qxf4
>10550 <first :      4 64 10 12475  g4 !!
>10550 <first :      4 112 20 16188  g4 f5 exf6 Rxf6 Bd6
>10770 <first :      5 68 40 44395  g4 Qh3 Nd5 exd5 cxd5
>11210 <first :      6 67 80 130466  g4 Qh3 Nd5 exd5 Rd3 Qh4 cxd5
>12640 <first :      7 67 220 386080  g4 Qh3 Nd5 exd5 Rd3 Qh4 cxd5
>15600 <first :      8 73 520 1054513  g4 Qh3 Nd5 exd5 Rd3 Qh4 cxd5 Bb7
>23950 <first :      9 54 1360 2774173  g4 Qh3 Nd5 exd5 Rd3 Qh4 Bg3 Qg5 cxd5
>83160 <first :     10 51 7280 13997968  g4 f5 exf6 g6 Bg3 Qxf6 g5 Qxg5 Be2 Qh6
>193780 <first:     11 -4 18340 36003605  g4 Qh3 Nd5 Qf3 Rd3 Qe4 Rd2 exd5 Qxe4
>dxe4
>417160 <first:     11 -1 40680 83217254  Bg3 !!
>473020 <first:     11  9 46270 93970165  Bg3 Qxh5 Kh1 a4 f3 h6 Ne4 Qg6 Rfe1 Rfc8
>
>The score drops after about 3 minutes on my computer and Bg3 found after almost
>7 minutes, so with standard timecontrol, fast computer and hashdata the move
>would probably be different :o). Only a slight chance with 40/40' IMHO.

I think that a better time allocation can help to find the move.
My opinion is that gandalf does the same mistake that other programs do in time
allocation.

It needs more than 3 minutes to calculate the exact score of g4 but I guess that
it needs less than it to see that the move fail low.

g4 is a move that you can suspect that it is wrong because of king safety
problems so the fact that the main line begins with g4 should encourage the
program to use more time.

The fact that the score dropped with g4 should also encourage it to use more
time.

The evaluation showed the scores:
73 depth 8
54 depth 9
51 depth 10

The fact that the evaluation is positive and goes down should encourage the
program to use significantly more time.

It is risky to use significantly more time in this case when you have not many
move to the time control but it is not risky when you have 25 moves to the time
control.

Using twice than the regular time even before seeing the fail low at depth 11 in
this situation is not risky and the regular time is clearly more than 1 minute
because part of the moves were in book.

After seeing the fail low it is clear that it should use more time.

Uri



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.