Le 3 sept. 2010 à 10:26, Fulvio a écrit :
> Joost 't Hart wrote:
>>
>> On 09/01/2010 06:11 PM, Fulvio wrote:
>>
>> Hi,
>>
>>
>>> There are some aspects that i want to change in scid:
>>>
>>> In the Tree Window:
>>> 1) I want to change the the performance statistic. Actually it is just a
>>> lookout in a table based on the opponents elo average, i want to use the
>>> fide formula for the performance calculation and express the result as
>>> +/- over the AvElo (es instead of the actual AvElo 2700 Perf 2695 i'll
>>> prefer AvElo 2700 Perf -5)
>>>
>>>
>>
>> Calculation is fine for me; I do not share your preference for
>> displaying, which I feel is rather unusual.
>>
>
> Just to clarify my idea: the actual perfomance calculation don't take in
> consideration the average elo of the player.
> If a 2700 player beats a 2500 player you get a performance of 3499 (!?)
> If a 2000 player beats a 2500 player you get a perfomance of 3499
>
> I remember Nunn saying in "Secret of Practical Chess" that the perfomance is
> significative only compared to the average elo, and i argree.
> If there is a line with AvElo 2500 and Perf 2600 i can consider it a very
> good line.
> Another line with AvElo 2700 and Perf 2650 is to be considered inferior.
As a user searching for promising new opening lines, I can only agree with
Fulvio.
I'll think the same of the above lines he cited.
> What i'm thinking about is calculating the FIDE expected result { 1 / ( 1 +
> pow10((oppElo - elo) /400.0)) }, calculate the perfomance accordingly and
> show the difference from the AvElo.
> This way a + line is a good line (players who played that move performed
> above their elo) and minus line are dubious (they lose elo points).
Btw, there is also another thing that would be great to have an idea on the
playability of an opening line: given a selected list of games where the
opening is played (say by players with > 2500 elo),
it would be great to be able to make the engine evaluate the position of all
games in the list after move X (X would be a parameter typically between 12
and 20) and report the average score it gets for these games. analyzing the
opening in this way would remove the noise in the performance due to inferior
moves played after the opening phase (or close to the time control for instance
;-). This is particularly important for variants that are not that much played
or rather new (hence have few games recorded).
Of course, the analysis could take some time for a large game list, but then we
could tune another parameter to fasten the computation (if desired): the depth
at which the engine has to go to evaluate the position after move X.
I've got no idea if this is complicated to implement, but it seems all elements
for such a computation are already in SCID. And I believe that would be much
helpful, don't you think?
Vincent
------------------------------------------------------------------------------
This SF.net Dev2Dev email is sponsored by:
Show off your parallel programming skills.
Enter the Intel(R) Threading Challenge 2010.
http://p.sf.net/sfu/intel-thread-sfd
_______________________________________________
Scid-users mailing list
Scid-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/scid-users
