Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Elo-related question, how to rate puzzles?

Author: Andrei P

Date: 11:31:39 12/02/04

Go up one level in this thread


On December 02, 2004 at 08:23:22, Drexel,Michael wrote:

>The outcome of such a 100 games match between lets say a 2200 rated player and a
>2300 rated player is not at all determined.
>
>First of all ratings are very often not accurate.
>
>Even if we assume they are (both players achieved their ratings against the same
>opponents with a significant number of games) the higher rated player of course
>does not necessarily win such a match.
>
>For example:
>
>The 2300 player might be an agressive player and strong tactician with lots of
>risky and objective unsound variations in his repertoire.
>As a consequence he scores very well against <2100 rated oppponents.
>
>The 2200 player might prefer a calm positional style and play the opening
>accordingly.
>As a consequence he often has to be satisfied with draws against <2100
>opposition.
>
>Michael

Basically, I believe that if the puzzles are rated carefully against many
players than the puzzle ratings are very precise. so, if one runs them through
another batch of players, one would get precise average elo rating for this
group. But within the second group you will see fluctuations:tacticians will get
higher "puzzle" elo than their fide rating, the opposite being true for
positional players.

Rating puzzles should be very accurate if you use enough people-testers. It
should not be any worse than fide rating. in fact, it should be better because
puzzles are "stable" in their strength.

Yes, a weaker player can beat a stronger player in their playing styles are
different. But if you have a large pool of players all this averages out. That
is exactly the reason why the weaker player is 2100 and the stronger 2300.

So, with large pool of players-testers one should be able to rate puzzles very
precisely. Of course it will only show the "tactical" elo and the precise elo
points will be dependent on how many "tacticians" you had amoung your testers.
but, again, it should not be an issue unless one selects too few players or
biases them in some way. I don't believ Livshitz biased his selection. but it is
not clear how many player actually rated the puzzles.



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.