New idea or already known and used ?

[De Noose Daniel]

This night I had a dream ... Maybe it's not possible to implement this or maybe it's a bad idea. But the idea is to solve lot of problems/puzzles.

When an engine starts thinking, it uses this order :

1. First search : a triple nullmove search. If there is an interesting position arising, keep track of the 3 moves played.
2. Second search : a normal search but when a move searched is one from the tracked ones, put it on top of move order. For these moves don't do any pruning until depth 6 (for example). For the other moves, you can do any pruning you want.

The idea is to try to obtain a good position that uses the nullmove pattern.

I don't know if I explain clearly what I'm thinking about because my english isn't very good and I'm not a programmer.

[Dann Corbit]

I don't think it will help a lot in general because at the start of the game of chess, the good moves are well known, and chess engines already do a lot of pruning in general searching (e.g. the branching factor for Stockfish and other very strong engines is about 1.5).

But it could not hurt to try it.

[De Noose Daniel]

Thanks Dann.

I think about it more in the case of searching puzzles like in ACT-1B, TTT1, ... where it could be interesting to find THE idea of the position, THE key ...

[Dann Corbit]

I think the idea would be better for quiet moves than for test problems.

Lots of test problems involve either sacrifices or zugzwang positions. Both of these are problematic for null moves.
Often, giving up a piece looks worse than a pass.
And when you are zugzwang, you don't want to pass.

[De Noose Daniel]

Yes Dann, you're right. I badly explained myself : my idea is not really nullmove but only a nullmove for the opponent. As if you can play 3 moves and not the opponent. Maybe there is another name for this.

[Dann Corbit]

The reason null move works so well is that most moves are bad.
A move that is worse than doing nothing does not need to be explored as carefully most of the time.
But test suites often turn this sort of idea on it's head.
That is the reason that they are fun. They usually involve a surprising sequence.

[Dann Corbit]

Yes Dann, you're right. I badly explained myself : my idea is not really nullmove but only a nullmove for the opponent. As if you can play 3 moves and not the opponent. Maybe there is another name for this.

I don't understand how this will help.
Everything will look winning.

[De Noose Daniel]

Yes Dann. As usual I badly explained : by interesting position I wanted to say a mate score ...

[Dann Corbit]

Yes Dann. As usual I badly explained : by interesting position I wanted to say a mate score ...

OK, so you want to exploit a known score. For example, "Here is a collection of 100 mate in 6 problems."

I think that idea (exploiting a known score) has merit, but the way that I would exploit it is with zero window searches.

If you know that the score is always mate in 6, then set a zero window around that score and search all candidate moves.
I think that would solve the whole set very quickly.

I do not know if it would work to find shorter solutions (except by accident) however. Another interesting aspect of test sets is bugs in them.
It is even more fun to find a shorter solution or a refutation.

[De Noose Daniel]

The idea is not searching a mate when you know there is one.

I'll try to explain more clearly.

1. you first make a search where you play 1-2-3 moves without the opponent can.
2. if you get a mate score, you keep track of all the 1-2-3 moves you play.
3. you make a normal search (as usual) but :
- each time you are searching one of the tracked moves, you give a threaten bonus to that variation (and you can try not to make pruning on these variations)
- you change your move order to put these variation on top in the next depth search.
I hope it's more clear.

Sorry again.