Cache, Pruning, Nullmove and Alpha-Beta

[Reinhard Scharnagl]

Hi all,

because of my latest considerations I am actually convinced, that most common pruning effects will depend on the alpha and beta limits under which they have been produced.

Caching upper or lower bounds into and from a transposition table thus sometimes will become errornous, if the actual pruning situation would have been changed, because alpha and beta in fact have altered.

My conclusion from this is that pruning should be independent from alpha and beta bounds. Thus the nullmove heuristic seems to be problematic too (I still do not use it).

What is your opinion to that observation? Am I wrong here?

Regards, Reinhard.

[Uri Blass]

My opinion is that pruning should be dependent on alpha and beta and I have more pruning that is dependent on alpha and beta that is not null move pruning and I will probably have more pruning in the future.

I think that not trying to use knowledge to decide if to prune is a mistake and alpha and beta is part of the knowledge.

Uri

[Reinhard Scharnagl]

Hi Uri,

My opinion is that pruning should be dependent on alpha and beta ...

well, that alone is of course ok, but my question is targeting on prosumed problems when then storing new bound results in a transposition table and using them in a changed alpha-beta context.

Reinhard.

[Uri Blass]

Hi Reinhard,

I think that using hash for pruning is problematic for the same reason and I still do not use it for pruning now.

I believe that it is possible to do it in a productive way(for example mate scores can be used for pruning) but a lot of improvements are possible without it and I prefer to try to improve
Movei in that direction.

In my case the problem is not only pruning but also that the evaluation is path dependent.

Uri