jhaglund wrote:
The only thing weak is the lack of hdd space & memory for file I/O. The databases generated would be any specified variation line analyzed up to x.
I have it on my long life list of goals to find someone to code this into Crafty, preferably, because I want to see it be #1.
Even along the principal variation (PV), it is already quite risky at depth x = 3 from the node you are analyzing.
x = 1 yields an ordinary opening book as if from Crafty games.
x = 2 yields a book with more variety but loses some quality
x = 3 is risky
x = 4 is a minor mine field
Don't trust even PV moves beyond x = 3 (except there is a mate).
Even along the principal variation (PV), it is already quite risky at depth x = 3 from the node you are analyzing.
x = 1 yields an ordinary opening book as if from Crafty games.
x = 2 yields a book with more variety but loses some quality
x = 3 is risky
x = 4 is a minor mine field
Don't trust even PV moves beyond x = 3 (except there is a mate).
Matthias.
The part you didn't know is x has to have an evaluation range +/- .25 of current position to be stored.
It's also why I wanted position learning reimplemented into Crafty.