Lazy Eval:
How many of you guys use it? How much of a speed-up did you get? What do you generally think about it?
I?m doing some experimenting with a very simple implementation, that is, I compute material balance with piece-square tables and check the value against alpha and beta using a defined margin. On Q-Search only.
So far the results haven?t been so clear. Setting a too small safe margin makes the program faster, but it plays worse.
Apparently, for my current evaluation function, 225 (centipawns) is the safest value that still gets me some speed up, namely 4%. With 100 I?ve got 7%, but the version got worse scores on EPD suites than the non-LE version.
In games between versions, LE version seems to do better, but barely.
So I?m not sure if the benefits are worth the risk of tactical blunders.