BB+ wrote:Does CCRL still do the "engine distance table"? This was
asserted as evidence that Rybka was not that close to Fruit (in the face of one of Diepeveen's posts) back in Feb 2006 when this topic was hot. Does anyone know exactly what this list was? Kirill Kryukov apologised for a lack of explanation (at that time) of the correlation numbers and said:
The good point of those correlation stats is that they show objective comparison of engine playing style, derived simply from their games. So it does not depend on disassembling and any subjective observations, but only on the evaulation and moves in the games.
It can be found on the 40/40 page. Look for the 'Correlation' link at the top of the list.
Here is the link to the Correlation page:
http://www.computerchess.org.uk/ccrl/40 ... ation.html
The data collected is ponder hits ( prediction of the other engine's move ) and position evaluation.
When looking at the ponder hit data for engines of different families, some interesting things can be seen.
Such as the relationship of other engines to Fruit, as well as other engines to Rybka. The same results
appear when analyzing data generated by Don's similarity program.
I am not as certain as to the utility of comparing positional evaluations ( you can judge that much better than I can ),
but a connection between Rybka and Fruit ( and/or Togas ) can be seen.