Author: Kirill Kryukov
Date: 08:49:42 02/25/06
Go up one level in this thread
On February 25, 2006 at 11:25:39, Sergei S. Markoff wrote: >Hmm, I think it must be clear. At the top of your table I see: "The upper one is >the evaluation difference, in pawns". So, I mean the pair correlation, not the >average difference of evals. For example, someone want to create a clone simply >changine engine output. For example you can multiply engine eval on 1.5, than it >will play the near same strength, but your criterial will show difference. But >when you're using the pair correlation it will be clear that it's a clone. > >Here you can found a formulas and several criteria description: > >http://www.3ka.mipt.ru/vlib/books/Programming/ComputerScience/Numerical_Recipes_in_C/c14-5.pdf I see what you mean. Currently I don't plan to compute pair correlation, because theoretically an engine can print one evaluation and use completely different eval internally. The formula for computing printed eval from internal may be very complex, and a motivated cloner is no doubt capable of doing that. So the low pair correlation is still not a proof that an engine is not a clone. Well, it should be easy to do thoigh, so I guess I may do this when I have time. IMHO, the most reliable way to spot engine similarities is still the ponder hit statistics, because it shows how similar is what two engines actually do on board, while any evaluation they print is just numbers, they can print whatever they like. Best, Kirill
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.