Why not two consecutive null moves ?

[mar]

Interesting. I recently switched from NMP+verification to hybrid NMP+NMR, this seems to have helped a bit in some positions while not losing elo (bullet only).
i.e. at low depths I prune (<=6) otherwise I reduce current node to 2/3 depth

[hgm]

This reduced search is just to check if there is a zugzwang? In that case it seems a wast of time when you have many mobile pieces. What do you do if the reduced search fails low? Do a full-depth search?

[mar]

Yes I do (in parent node). To be more specific, I do this:
do nullmove, if it fails high, then I either prune at low depth or reduce current node (expecting to fail high).

[Henk]

Yes I do (in parent node). To be more specific, I do this:
do nullmove, if it fails high, then I either prune at low depth or reduce current node (expecting to fail high).

So what happens in the case you reduce current node and it fails low ?

[mar]

It almost never fails low.
If a zw node fails low then it produces a cutoff in parent node.

[mar]

Sorry, of course I don't because it then produces a cutoff.

[bob]

How many real plies does a null move reduction need to make it work correctly ? For instance do I need to test depth >= R + 2 if Null move reduction reduces with R. (My null move reduction does not reduce with R + 1)

Code: Select all

 if &#40;depth >= R + 2 && depth < pos.MaxDepth&#41;
 &#123;
        if &#40;NullMoveCutOff&#40;entry, depth, ub, out nullMoveValue&#41;)
        &#123;
             bestMove = null;
             return nullMoveValue;
        &#125;

 &#125;

The main danger is allowing a null-move to take you to a VERY selective search where you miss tactical threats, particularly mates. If your q-search includes at least one ply of non-capturing checks, then you can feel pretty safe reducing with r=3 everywhere, since you still get checks even if you collapse directly into q-search. If you do not do q-search checks, I would leave one ply of full-width search on the end to be sure that you are not playing oblivious to a serious tactical threat.

[Henk]

How many real plies does a null move reduction need to make it work correctly ? For instance do I need to test depth >= R + 2 if Null move reduction reduces with R. (My null move reduction does not reduce with R + 1)

Code: Select all

 if &#40;depth >= R + 2 && depth < pos.MaxDepth&#41;
 &#123;
        if &#40;NullMoveCutOff&#40;entry, depth, ub, out nullMoveValue&#41;)
        &#123;
             bestMove = null;
             return nullMoveValue;
        &#125;

 &#125;

The main danger is allowing a null-move to take you to a VERY selective search where you miss tactical threats, particularly mates. If your q-search includes at least one ply of non-capturing checks, then you can feel pretty safe reducing with r=3 everywhere, since you still get checks even if you collapse directly into q-search. If you do not do q-search checks, I would leave one ply of full-width search on the end to be sure that you are not playing oblivious to a serious tactical threat.

My q-search does not recognize mate. It only searches zero or more captures. So if it gives the opponent a full-width search returning a move that attacks a king it would not be detected in q-search. It would only prevent it from doing captures (except capture of the king) at level 0. That means that two full-width searches looks safer. But maybe that might be too expensive.

[bob]

How many real plies does a null move reduction need to make it work correctly ? For instance do I need to test depth >= R + 2 if Null move reduction reduces with R. (My null move reduction does not reduce with R + 1)

Code: Select all

 if &#40;depth >= R + 2 && depth < pos.MaxDepth&#41;
 &#123;
        if &#40;NullMoveCutOff&#40;entry, depth, ub, out nullMoveValue&#41;)
        &#123;
             bestMove = null;
             return nullMoveValue;
        &#125;

 &#125;

The main danger is allowing a null-move to take you to a VERY selective search where you miss tactical threats, particularly mates. If your q-search includes at least one ply of non-capturing checks, then you can feel pretty safe reducing with r=3 everywhere, since you still get checks even if you collapse directly into q-search. If you do not do q-search checks, I would leave one ply of full-width search on the end to be sure that you are not playing oblivious to a serious tactical threat.

My q-search does not recognize mate. It only searches zero or more captures. So if it gives the opponent a full-width search returning a move that attacks a king it would not be detected in q-search. It would only prevent it from doing captures (except capture of the king) at level 0. That means that two full-width searches looks safer. But maybe that might be too expensive.

Just so you have at least one ply to recognize mates. Which means you have to extend when you give a check rather than extending when you escape check, or you will need 2 plies at the end.