Question Regarding PV-Collection

[StackFish5]

Currently in my chess engine, I use a pv-collection system:

Code: Select all

MOVE pv_line[MAX_PLY];
...

//PV Collection in alpha-beta
if(score > alpha){
	alpha = score;
	pv_line[current_ply] = move;
}

However, I'm wondering whether about the benefits triangular pv-tables are worth it as it seems that many people use it. To me, it appears to be overly redundant and slow, especially the part were the pv move is copied to all other occurrences applicable. Are there any benefits to using triangular pv-tables?

[abulmo2]

The triangular PV gives a correct PV, your PV collection system returns a random sequence of moves, but not the PV.

[AAce3]

Currently in my chess engine, I use a pv-collection system:

Code: Select all

MOVE pv_line[MAX_PLY];
...

//PV Collection in alpha-beta
if(score > alpha){
	alpha = score;
	pv_line[current_ply] = move;
}

However, I'm wondering whether about the benefits triangular pv-tables are worth it as it seems that many people use it. To me, it appears to be overly redundant and slow, especially the part were the pv move is copied to all other occurrences applicable. Are there any benefits to using triangular pv-tables?

This doesn't look correct.
If you have a tree, the first best move will be correct. The second best move will be whatever raises alpha in some other branch, and the third move will be some other move that raises in some other branch, etc. etc. Basically, you have no way of ensuring that the pv moves are all from the same branch of the tree.

[flok]

Currently in my chess engine, I use a pv-collection system:

Just curious: but why not just walk the TT?

[hgm]

Currently in my chess engine, I use a pv-collection system:

Just curious: but why not just walk the TT?

Because the TT often gets overwritten. Joker retrieved the PV from the TT after the search, but often the PV ended in a position that did not have the same evaluation as the move that was played. (Which was especially obvious when checkmates were involved.)