TT: depth field

[flok]

Usually one stores the search-depth in de TT-depth field.
I was curious if an alternative would work.
Did not want to change the structure too much.
So I tried storing the number of nodes searched in a subtree; the log2() of it that is.
The result is -117 elo.
Maybe this inspires someone.

[AndrewGrant]

Probably can't work, because node count may be more of a marker of how wide, or how wrong the [alpha, beta] window was, than the actual depth of the search. Interesting nonetheless.

[flok]

Probably can't work, because node count may be more of a marker of how wide, or how wrong the [alpha, beta] window was, than the actual depth of the search. Interesting nonetheless.

That was exactly my goal: as an indication of how "thorough the search of that subtree" was.

[abulmo2]

I did that in my Othello program Edax. It helped to reduce the size of the searched tree. I built a draft from the age of the search, its cost (log2(node count)), the selectivity level and the depth, to choose which entry to replace.

[Steve Maughan]

I've thought about adding this type of approach. However, in a Lazy SMP environment there could be a quick cutoff due to another thread, which invalidates the whole approach.

— Steve