(possible stupid question) asymmetrical search

[xr_a_y]

I was wondering if anyone already tried with success a kind a asymmetrical search. By this I mean a search where the (root) side-to-move will (for example) prune a lot but the opponent turn will be search more carefully.
Does that make any sense ?

[Michael Sherwin]

Yes, it makes sense in a way but it depends what it is that you are trying to achieve. If your aim is to create an interesting opponent for a human this is a great idea because it would be more vulnerable to human tactics. If you want to contest against fish and you want to use that idea then you can do two searches one with the root side like you said and the other with the opponent's side. Then the scores can be compared. If the scores are close there is most likely little danger in the position. If there is a big difference in the scores then there are dangers for one side or the other. Of course both searches being greatly reduced would allow both to look deeper in less time.

[bob]

Richard Lang did this in the old Chess Genius engines. As but one example, you can make it harder to make a tactical mistake yourself, using caution, but then make it harder to find wild tactical shots since you don't do the same types of things for both sides. Idea made sense back in the days of limited search depth. Not sure it makes much sense today, however.

[xr_a_y]

I ask having something about king safety in my head. I think it is not always easy, even with asymetrical evaluation, to handle something like :
"I dont do crazy sacrify when attacking, but the opponent may do it", or the other way around ; because this concern not only evaluation but also pruning margins. Another thing is, if I'm ahead, I want to look for mate that involve giving some pieces, so I don't wanna prune too much anymore but the opponent will still search as usual.
So mainly when there is a material imbalance or many piece targeting a king, I want to try asymetrical search ...

[Michael Sherwin]

The trouble for doing it only for one side is it is not known ahead of time where the opportunity for attack or defense exist. If you reduce the critical move then the solution will be missed. You might find a brilliant win or you may suffer a short sighted loss.

So let's assume the case where no win is found. Then an asymmetrical search should then also be done for the opponent's side to judge the danger to the engine. After the second search if there are two best moves to choose from you can decide which move to play. Or you may then choose to spend more time to search each best move using symmetrical search to decide between which move is played.

[Michael Sherwin]

As a related or as a helper notion I always thought it would be a good idea to count the number of alternative good moves there are at early ply in the search. Let's say at the third ply there are more than n good moves. Then the engine can enter into a asymmetrical search because there are alternative moves to play. Or maybe the overall search depth can be reduced. If several ply in a row have at least n good moves then the search may be able to be greatly reduced.