Author: Christophe Theron
Date: 17:32:31 09/02/01
Go up one level in this thread
On September 02, 2001 at 14:27:18, Chessfun wrote:
>On September 02, 2001 at 12:17:24, Torsten Schoop wrote:
>
>>Hi Uri,
>>
>>
>>>If it is 5% slower than it means that it seems to be 5 elo
>>>and my impression based on posts of amir in the past is that 5%
>>>is more realistic.
>>
>>
>>IMO it is not possible to measure a speed different of even 5% in ELO at 40/120
>>on 1 GHz machines (maybe 1 ELO?). For that reason it makes no difference to use
>>Junior 7 or Deep Junior 7.
>>
>>Torsten
>
>On my TB945 mhz from the start position
>Junior 7 655 kN/s
>Deep Junior 7 604 kN/s
>
>Though I don't see how 5% translates into only 5 elo.
>
>Sarah.
Here is the general formula for elo difference calculations based on speed
differences:
elo_diff = 70 * log(speed_ratio) / log(2)
(it does not matter which log you are using, base 2 or base 10)
In this case:
elo_diff = 70 * log(655/604) * log(2) = 8.19 elo
(if you calculate with 604/655 you get -8.19 elo, that's a property of the
formula).
So let's say 8 elo points of difference.
Now how many games do you need in order to be able to detect an 8 elo points
difference?
If you want only 70% confidence in the result, you must play 1400 games at
least.
If you want 90% of confidence, you must play more than 1600 games (much more,
actually I do not have it in my tables but it must be over 2000 games).
Good luck. :)
Christophe
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.