Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: The Insignificance of the CSS WM test of Meiler, Schumacher & Gurevich

Author: Bernhard Bauer

Date: 01:42:46 06/23/04

Go up one level in this thread


On June 23, 2004 at 03:42:41, Steve Glanzfeld wrote:

>On June 23, 2004 at 03:01:36, Bernhard Bauer wrote:
>
>>r= 2600 + 2*number_of_solved_pos_in_% - 5*solution_time_in_min/100
>>
>>could you explain the value of such a formula?
>>
>>Of course anything may have some value for somebody, so what?
>>My program, which plays up to now very poor chess has already achieved 2500
>>by the WM-test. Ha, ha!
>>The exel table is fantastic. So much work which makes so little sense!
>>Incredible!
>
>The value 2600 at the beginning of that formula has NO INFLUENCE on the
>rankings. It's just added for convenience obviously, to create ratings at a
>common level, for easier comparison. IMO the important part is
>
>2*number_of_solved_pos_in_% - 5*solution_time_in_min/100
>
>This just means that a maximum of 200 points can be reached by the solvings, but
>the time required to solve can reduce that value by up to -100 points (5*20
>min./pos.*100 pos./100). IOW., the percentage of solutions has 2/3 influence on
>the rating, and 1/3 is determined by the time consumption. This makes sense for
>a difficult test (only few engines solve 70 pos. or more). The solvings are the
>main thing, and in addition to that the times can provide a finer grid (i.e. to
>rank engines which have the same number of solutions).
>
>I don't understand your sarcasm about this... do you have a better rating method
>for tests? Btw. no, I say RATING (test rating), not Elo. Forget about Elo. Tell
>me your great universal formula which is beyond any critizism. :))
>
>Steve

While we agree that a set of positions can not give us information about playing
strength one number, kind a norm, may be helpfull in comparison.

A formula like this has been proposed earlier.

r = base_value + (max_value - base_value) / number_of_positions
    * sum( 0.5^(solution_time/If solution_time equals base_solution_time you add
(max_value - base_value)/2.))

by ^ I mean exponentiation.
This formula has some properties.
You may run your test as long as you want.
If solution_time equals base_solution_time you add (max_value - base_value)/2.
If solution_time equals 2*base_solution_time you add (max_value - base_value)/4.
For a harder test base_solution_time should be greater than for an easy test.
No nead to implement artifical values like LQ. If you have a solution time, you
have although solved that position.

regards
Bernhard




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.