FEN database (287 millions posiitions)

[Adam Hair]

I make use of Don Dailey's similarity tool and Michel van den Bergh's branch of Polyglot and wb2uci. The adapters are necessary to log the engine output. Miguel saved me the trouble of trying to write a script to extract the eval score at depth 'x' (I have collected data at depths 4, 5, 6, and 10 so far).

You have to learn how to "unwrap" the similarity utility so that you can insert epd files and modify the 'sim.tcl' file. After making the modifications, I rewrap the files into an executable, but Don says it is unnecessary.

It is easier to use Arena's automatic analysis, but it is much slower. I have gotten quite use to modifying the similarity tool to collect different sorts of data and find it quite handy to use.

An additional step I took was to find engines that are approximately the same strength as Gaviota or stronger at fixed depth. I use Komodo, Houdini, Toga, Nemo, Bright, Spark, Spike 1.2, Critter 0.9, and a few other engines. I do not know how Djinn compares at fixed depth, but I assume these engines would work for you also.