Did people try replacing LMR by pruning

[Michael Sherwin]

The idea is the following:

Every time that LMR conditions happen you do not change the depth but increase pruning variable by 1 when the pruning variable is simply the number of times when LMR conditions happened during the path(of course you need to reduce it when you take back the last move).

If the pruning variable is bigger or equal to the remaining depth
then you decide if to prune or not to prune based on evaluation or based on result of the qsearch.

Uri

A few years back I tried LMP in RomiChess iirc P3a. That version is the only version that ever won a WBEC test tournament. And it did it with a perfect score of 18-0! When I changed the LMR to 2 I gave up on LMP as a LMR of 2 was just as efficient with out as much risk. You BTW, participated in that discussion. LMP was also used in Glaurung 2.x! I have not looked to see if it is retained in Stockfish.

[metax]

Stockfish 1.6:

Code: Select all

// Move count based pruning
          if (   moveCount >= FutilityMoveCountMargin
              && ok_to_prune(pos, move, ss[ply].threatMove)
              && bestValue > value_mated_in(PLY_MAX))
              continue;

[Michael Sherwin]

Stockfish 1.6:

Code: Select all

// Move count based pruning
          if (   moveCount >= FutilityMoveCountMargin
              && ok_to_prune(pos, move, ss[ply].threatMove)
              && bestValue > value_mated_in(PLY_MAX))
              continue;

I am glad to see that the idea survives!

[jdart]

Toga used this general idea also - that was the first place I remember seeing it. But I have not found it useful, myself.

--Jon

[Michael Sherwin]

Toga used this general idea also - that was the first place I remember seeing it. But I have not found it useful, myself.

--Jon

I believe that my testing of it and posting about it precedes its use in Toga!