Identical positions in tree

[shawn]

Thank you very much everybody!! This helps!! Before TT the mate in 6 from #8 took my engine above 30 minutes, now 6 minutes.

One last(?) question:
At what depth do we decide to not at all use the transposition table?
At very low depth I see two reasons for not doing so:
- the number of positions that we would WANT to store grows exponentially
- doing the evaluation without TT becomes fast enough if not much faster
I am currently doing from depth 5 on but that is also a bit arbitrarily chosen.

You should do it for every depth, including within Quiescence search. TT probing is still very fast compared to searching and evaluating positions, and you could potentially use scores from higher depth entries to enhance the score used at those lower depth entries where TT cutoff is performed.

[hgm]

It is quite important to always store a result in the TT, even if this would overwrite a result from a much deeper search. If you want to protect the latter somewhat, because they represent more search effort, you should solely do it by deciding which of a number of entries you will overwrite. But you should not over-estimate the importance to hang on to deep results, because they typically have to be kept very long before an identical position is visited in the tree. A TT entry that saves you a lot of work only very infrequently is not much more useful than an entry that very frequently saves you a little work, bacause it was used to (sequentially) store many positions close to the leaves, which experience transpositions very shortly thereafter.