Depth reduction for moves that don't save the queen

[sedicla]

Hello,
I'd like to share this idea so maybe you guys have tried or would like to take a look.
When queen is attacked reduce the search of quiet moves, except queen moves of course. Also research if reduced search fails high. Seems promising since the research rate is very low.
If your queen is attacked don't need to worry too much about moves that don't save the queen
I'm still testing it, don't think is a lot of gain but maybe someone is already doing that or can take another look.
Another possibility is to do the same for rooks when queens are of the board, and so on for minor pieces.

[Evert]

Depends very much on how you implement this, I think.

"Move that doesn't save the queen" is too vague a concept to be useful. It either means very little ("any move that doesn't move the queen") or is very expensive to calculate ("any move that doesn't move the queen to a safe location, blocks the line of attack or captures the attacker", but the latter two won't work if the queen is attacked twice). There's also an assumption that the queen is the piece that is most likely to be in danger of being captured. I'm not sure that's actually true.

What I've tried is the following: if the null-move is refuted by a capture, then that capture represents a threat (and it should be the most dangerous one, assuming you have good move-ordering for captures). We can identify the threatened piece and the attacker and encourage moves that try to do something about the thread (move piece away, capture the attacker). So far, I haven't been able to make that work in terms of playing strength though...

[sedicla]

At this moment the reduction is for moves that don't move the queen. Seems expensive to calculate more than that now. I assume these moves that acually protect the queen like blocking the attacker will be then researched properly.
Also I like to point that I don't reduce if move is already lmr reduced.

[lucasart]

Hello,
I'd like to share this idea so maybe you guys have tried or would like to take a look.
When queen is attacked reduce the search of quiet moves, except queen moves of course. Also research if reduced search fails high. Seems promising since the research rate is very low.
If your queen is attacked don't need to worry too much about moves that don't save the queen
I'm still testing it, don't think is a lot of gain but maybe someone is already doing that or can take another look.
Another possibility is to do the same for rooks when queens are of the board, and so on for minor pieces.

I found an improvement in my program along those lines in my program. Basically I don't reduce moves that attack the queen (directly or indirectly) in a credible way (ie. moved piece is not attacked or is defended by a pawn and and not attacked by a pawn) when the remaining depth is low.

And *to my surprise*, such a horrible and hacky piece of code did produce a measurable elo improvement...

[bob]

Hello,
I'd like to share this idea so maybe you guys have tried or would like to take a look.
When queen is attacked reduce the search of quiet moves, except queen moves of course. Also research if reduced search fails high. Seems promising since the research rate is very low.
If your queen is attacked don't need to worry too much about moves that don't save the queen
I'm still testing it, don't think is a lot of gain but maybe someone is already doing that or can take another look.
Another possibility is to do the same for rooks when queens are of the board, and so on for minor pieces.

Here's a point to ponder. If you are sure a move is bad, reducing it will be helpful. For example, a queen capture of a lower-valued piece that is defended, or a queen capture that loses material according to SEE.

But reducing moves you are NOT sure about can be dangerous. For example, reducing all but queen moves when the queen is attacked is almost certainly going to be wrong. Most of the time, when the queen is attacked, you can just rip the attacking piece and end the threat instantly. If you reduce the move that captures the threat, then your search may think that the only viable choices are moves that actually move the queen, which can cause misleading evaluations. Ditto for interpositions. A common move, like Bg5 attacking black's queen at d8 but where black can play f6 blocking the attack AND requiring that white waste a move retreating the bishop is worth looking at carefully. Reducing it can produce the wrong score...

We always have to deal with errors in such decisions, but given the choice of reducing something important or not reducing something irrelevant, I'd prefer the latter, as it will slow you down, but won't cause incorrect evaluations. The former can be ruinous...

[Sven]

If you reduce the move that captures the threat

He wants to reduce quiet moves only.

EDIT: interpositions should perhaps be excluded from reductions in this case since they might also save the queen.

Sven

[bob]

If you reduce the move that captures the threat

He wants to reduce quiet moves only.

Sven

I realize that, but I wanted to make a clear point. I think it perfectly safe/reasonable to reduce bad captures that don't meet some SEE criteria. And the interpositions are also important, yet would get reduced..