LMR: history or not?

[Alessandro Scotti]

Hi,
anyone tried LMR not based on history? In a recent test Hamsters scored 471/800 with LMR+history and 469/800 with same LMR but no history. In other words, if the LMR prerequisites are satisfied (not in check, not an "interesting" move and so on) the move is reduced regardless of its history. I will run another test but so far it seems the two are practically equivalent. However, the "historyless" version is simpler and more appealing to me... have you tried this?

[hgm]

I think Crafty has stopped using history information for the LMR decision (or in fact for anything) a long time ago. Micro-Max does not use history either. The only moves exempted from LMR are the hash move, check evasions, all captures and all Pawn pushes there. Joker also doesn't account history. It does exempt moves from LMR that are captures, check evasion, passer pushes, or have had a score above alpha on any previous IID iteration.

[Vempele]

I think Crafty has stopped using history information for the LMR decision (or in fact for anything) a long time ago. Micro-Max does not use history either. The only moves exempted from LMR are the hash move, check evasions, all captures and all Pawn pushes there. Joker also doesn't account history. It does exempt moves from LMR that are captures, check evasion, passer pushes, or have had a score above alpha on any previous IID iteration.

So you reduce non-capture checks? In order to compensate for not reducing the replies (and thus practically extending the check by more than a ply)?

[hgm]

Micro-Max does reduce non-capture checks, for the simple reason that it has no idea if a move delivers check or not.

In Joker I don't know by heart anymore how it treats checks. They might even be extended (outside QS), and that would exempt them from LMR too.

[Alessandro Scotti]

I think Crafty has stopped using history information for the LMR decision (or in fact for anything) a long time ago. Micro-Max does not use history either. The only moves exempted from LMR are the hash move, check evasions, all captures and all Pawn pushes there. Joker also doesn't account history. It does exempt moves from LMR that are captures, check evasion, passer pushes, or have had a score above alpha on any previous IID iteration.

So history is not used anymore also for move ordering, correct? Is there any available test data that shows the difference of using/not using history in LMR and/or move ordering? If not, I can try to run a test myself but it will take quite a lot of time...

[bob]

Hi,
anyone tried LMR not based on history? In a recent test Hamsters scored 471/800 with LMR+history and 469/800 with same LMR but no history. In other words, if the LMR prerequisites are satisfied (not in check, not an "interesting" move and so on) the move is reduced regardless of its history. I will run another test but so far it seems the two are practically equivalent. However, the "historyless" version is simpler and more appealing to me... have you tried this?

I've not used history in a year now... I didn't find it helped at all...

[bob]

I think Crafty has stopped using history information for the LMR decision (or in fact for anything) a long time ago. Micro-Max does not use history either. The only moves exempted from LMR are the hash move, check evasions, all captures and all Pawn pushes there. Joker also doesn't account history. It does exempt moves from LMR that are captures, check evasion, passer pushes, or have had a score above alpha on any previous IID iteration.

So history is not used anymore also for move ordering, correct? Is there any available test data that shows the difference of using/not using history in LMR and/or move ordering? If not, I can try to run a test myself but it will take quite a lot of time...

I don't have any history counters in Crafty. I found that at the depths I see today, the history counters were simply saturated with nonsense...

[Alessandro Scotti]

I don't have any history counters in Crafty. I found that at the depths I see today, the history counters were simply saturated with nonsense...

Hi Bob,
did you just remove history from move ordering or tried to replace it with something else (e.g. difference between "from" and "to" in the piece/square table)?

[Harald Johnsen]

I made the same test and saw that history counter did not improve anything for the lmr choice. So the last cyrano (0.2d) is only doing a TT check (à la rebel) to decide to reduce or not.

HJ.

[bob]

I don't have any history counters in Crafty. I found that at the depths I see today, the history counters were simply saturated with nonsense...

Hi Bob,
did you just remove history from move ordering or tried to replace it with something else (e.g. difference between "from" and "to" in the piece/square table)?

I've not tried to replace it with anything... About the only ordering I do beyond the usual captures/killers/etc is that I generate moves in an order that moves toward the opponent first, then away later. But that's it to date.