On Sun, May 9, 2010 at 2:55 PM, Greg Schmidt <[email protected]> wrote:
> >* When assigning the credit value for a node which is a transposition, > are *all* parents which point to that node credited, or just the particular > one which led to that transposition in the current continuation? If the > former case, that's seems like the TT would lead to more efficient learning > (a bit "RAVE like" since a single trial results in multiple updates) > although wouldn't it be unwieldy for a node to have a list of pointers back > to its multiple parents? > * > >Obviously there's some overhead. My hash-table doesn't store positions, > but lists of tree->nodes that have the same hash-code. So you don't need to > keep lists of parents >explicitly, it follows from being in the same list. > > Yes, that' how it's typically done, but I don't see how that leads to > finding *all* parents of a given node. The situation I'm referring to is > when you're propagating the UCT value back up the tree following a > simulation. If you also want to update *all* parents of that node (not just > the single parent within the simulated line of play), how would you find > *all* of them? The parent nodes will have different hash codes (and be in > different chained lists) in your representation. Therefore I don't see > how this representation is able answer the question "give me the list of all > parent nodes which point to this TT entry" (unless I'm missing something). > To make this work I think you need to have a list of parent nodes contained in the record. It could of course be a linked list but each node would need it's own separate parent list if I'm thinking about this correctly. And a regular hash table does not solve this problem. So it's possible to do all of this more or less "correctly" but I guess it's an open question whether or not it's worth it. Maybe it is, I just don't know. > > -- Greg > > > > _______________________________________________ > Computer-go mailing list > [email protected] > http://dvandva.org/cgi-bin/mailman/listinfo/computer-go >
_______________________________________________ Computer-go mailing list [email protected] http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
