Author: William Penn
Date: 05:41:57 12/13/05
Go up one level in this thread
On December 13, 2005 at 01:52:22, Kurt Utzinger wrote:
>On December 13, 2005 at 00:10:16, William Penn wrote:
>
>[...]
>Also the length of analysis has been increased. The beta version
>>produces analysis displays about 10 ply long, which decreased to 2-3 ply at
>>longer times (several hours). In contrast the length of displayed analysis in
>>the Preview version starts at about 10-12 ply, and I'm seeing 14 ply after
>>letting it run for an hour in infinite mode.
>[...]
>
> Hi William
> Is this relevant? I have learnt to never trust
> the moves displayed after the first move of
> white an black. Have you ever checked how many
> times computers are changing their minds if
> you continue in the analysis mode the main
> move suggested?
> Best regards
> Kurt
It is relevant for me because that is the ONLY way I use chess software - with
very long analysis in infinite mode - a common practice by correspondence chess
players. If it can be shown that I am wasting my time thereby, I will stop
immediately(!), but hopefully the longer it thinks the better the choice of
move. In theory, letting it run for 2 hours on a 2 GHz processor is equivalent
to letting it run for 4 hours on a 1 GHz processor, or something similar. In
other words we can get a better analysis either by using a faster processor (or
some other hardware improvement) or by letting it run for longer times.
I must backtrack on my prior comment. The Rybka 1.0 Preview version still shows
a decreasing output with longer analysis. It still decreases to 2 ply at long
times. For example:
Engine: Rybka 1.0 Preview 32-bit (704 MB) by Vasik Rajlich (in Shredder Classic
9.1 GUI)
11.01 0:01 +0.47 26.Rg3 Kh8 27.Nxc5 b5 28.Kg2 Qd8
29.Rg1 Qd6 30.Ndb3 Rg8 31.Nd4 (110.827) 69
12.01 0:02 +0.46 26.Rg3 Kh8 27.Nxc5 b5 28.Kg2 Qd8
29.Rg1 Qd6 30.Ndb3 Rg8 31.Nd4 (193.266) 70
13.01 0:03 +0.43 26.Rg3 Kh8 27.Nxc5 b5 28.Kg2 Qd8
29.Rg1 Qd6 30.Ndb3 Rg8 31.Nd4 (265.088) 68
14.01 0:42 +0.48 26.Rg3 Kh8 27.Nxc5 b5 28.Kg2 Qd8
29.Rg1 Qd6 30.Ndb3 Rg8 31.Nd4 (1.807.218) 44
15.01 4:15 +0.45 26.Rg3 Kh8 27.Nxc5 b5 28.Kh1 Qb6
29.Ndb3 Qd6 30.Rbg1 Rg8 31.Bh5 b4 (10.076.211) 40
16.01 10:27 +0.45 26.Rg3 Kh8 27.Nxc5 b5 28.Kh1 Bf6
29.Rbg1 Qb6 30.Qc1 Bg5 31.Rxg5 hxg5
32.Rxg5 Rg8 (24.859.478) 40
17.01 28:58 +0.50 26.Rg3 Kh8 27.Nxc5 b5 28.Kh1 Bf6
29.Rbg1 Qb6 30.Qc1 Bg5 31.Rxg5 hxg5
32.Rxg5 Rg8 (68.172.709) 40
18.01 63:59 +0.46 26.Rg3 Kh8 27.Nxc5 b5 28.Kh1 Bf6
29.Rbg1 Qb6 30.Qc1 Bg5 31.Rxg5 hxg5
32.Rxg5 Rg8 (146.333.789) 39
19.01 166:03 +0.45 26.Rg3 Kh8 27.Nxc5 b5 28.Kh1 Bf6
29.Rbg1 Qb6 30.Bh5 b4 31.Ndb3 Bh4 (383.527.063) 39
20.01 443:48 +0.46 26.Rg3 Kh8 (1.049.074.071) 40
Someone may ask, "What is the position for this analysis?", but the position
doesn't matter. All infinite analyses do this after several hours. I expect this
will be fixed in a subsequent patch, as promised by the author in an earlier
thread here.
However, as you suggest, perhaps it remains to be proved whether there is
actually any improvement in the analysis and choice of best move at such long
run times - or is it just a total waste of time & effort?!?!?!
WP
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.