iterative deepening & a/b pruning
Posted:
22 Jul 2013, 07:25by Folkert van Heusden
Hi,
When performing iterative deepening, why shouldn't I re-use the calculated alpha/beta of the previous iteration?
I googled this and found some people saying "don't do this, use -inf/+inf" but I don't understand why.
regards
Re: iterative deepening & a/b pruning
Posted:
23 Jul 2013, 05:01by Ron Murawski
Hi Folkert,
When you finish a search to an arbitrary depth, you have no alpha/beta bounds, you have an exact value -- let's call this value 'score'. I use (score-margin) as the new alpha and (score+margin) as the new beta. where 'margin' is somewhere between 15 to 50 centipawns.
Ron
Re: iterative deepening & a/b pruning
Posted:
23 Jul 2013, 12:47by Folkert van Heusden
Ron Murawski wrote:When you finish a search to an arbitrary depth, you have no alpha/beta bounds, you have an exact value -- let's call this value 'score'. I use (score-margin) as the new alpha and (score+margin) as the new beta. where 'margin' is somewhere between 15 to 50 centipawns.
Ah indeed. I must experiment with this. Thanks!