Killer and History: Increased Node Count
Posted: Tue Jan 15, 2013 4:10 am
Hi,
I decided to revisit my move ordering and noticed something interesting. Both the killer history and move history cut many nodes when tested individually but when both heuristics are enabled in the search at the same time, the node count is slightly higher than the history by itself.
I'm not really sure why this happens. I use the standard History[64,64] for history and tested various killer arrays with the current KillerA[MaxPly] and KillerB[MaxPly] and I keep a counter for each to help with ordering and replacement.
I am sorting moves:
1. PV from previous iteration.
2. Position's move from the TT.
3. Captures (MVV/LVA) - no SEE.
4. Killer A or B based on which one has higher count.
5. Moves from the history.
Has anyone else experienced a similar situation?
I can only speculate that a killer move is being played but is not cutting and leading to more nodes tested; however, I would expect it would still decrease node count a bit when coupled with the history heuristic.
Thanks,
Cheney
I decided to revisit my move ordering and noticed something interesting. Both the killer history and move history cut many nodes when tested individually but when both heuristics are enabled in the search at the same time, the node count is slightly higher than the history by itself.
I'm not really sure why this happens. I use the standard History[64,64] for history and tested various killer arrays with the current KillerA[MaxPly] and KillerB[MaxPly] and I keep a counter for each to help with ordering and replacement.
I am sorting moves:
1. PV from previous iteration.
2. Position's move from the TT.
3. Captures (MVV/LVA) - no SEE.
4. Killer A or B based on which one has higher count.
5. Moves from the history.
Has anyone else experienced a similar situation?
I can only speculate that a killer move is being played but is not cutting and leading to more nodes tested; however, I would expect it would still decrease node count a bit when coupled with the history heuristic.
Thanks,
Cheney