lmr at PV nodes?

[silentshark]

Most implementations seem to avoid.. has anyone tried permitting LMR on PV nodes? If so, what were the results?

fyi, my quick testing seems to show a slight reduction in ELO if lmr is done on PV nodes, but margin of error is still high

[bob]

Most implementations seem to avoid.. has anyone tried permitting LMR on PV nodes? If so, what were the results?

fyi, my quick testing seems to show a slight reduction in ELO if lmr is done on PV nodes, but margin of error is still high

I do it everywhere, just like null-move, and have found that trying to restrict it always hurts in my testing, however I try to do the restriction.

Seems to me that if a move is bad and deserves reducing, it deserves the same treatment everywhere.

[silentshark]

Most implementations seem to avoid.. has anyone tried permitting LMR on PV nodes? If so, what were the results?

fyi, my quick testing seems to show a slight reduction in ELO if lmr is done on PV nodes, but margin of error is still high

I do it everywhere, just like null-move, and have found that trying to restrict it always hurts in my testing, however I try to do the restriction.

Seems to me that if a move is bad and deserves reducing, it deserves the same treatment everywhere.

that's really interesting - I like the look of the speedup doing lmr everywhere gives, but I'm not seeing any elo gain (or loss) yet. More testgames needed..

[bob]

Most implementations seem to avoid.. has anyone tried permitting LMR on PV nodes? If so, what were the results?

fyi, my quick testing seems to show a slight reduction in ELO if lmr is done on PV nodes, but margin of error is still high

I do it everywhere, just like null-move, and have found that trying to restrict it always hurts in my testing, however I try to do the restriction.

Seems to me that if a move is bad and deserves reducing, it deserves the same treatment everywhere.

that's really interesting - I like the look of the speedup doing lmr everywhere gives, but I'm not seeing any elo gain (or loss) yet. More testgames needed..

I believe you are going to need in the 100K+ game range to measure the difference. It is not very big.

[Mincho Georgiev]

I spent a month in testing some time ago and the results in my implementation was really bad when LMR and nullmove was restricted at pv nodes. The only exception i had was FP - 5 elo better when restricted.

[Greg Strong]

Most implementations seem to avoid.. has anyone tried permitting LMR on PV nodes? If so, what were the results?

fyi, my quick testing seems to show a slight reduction in ELO if lmr is done on PV nodes, but margin of error is still high

My experience showed a definite bonus if you do LMR at all nodes, including at the root.

[zamar]

Most implementations seem to avoid.. has anyone tried permitting LMR on PV nodes? If so, what were the results?

fyi, my quick testing seems to show a slight reduction in ELO if lmr is done on PV nodes, but margin of error is still high

For Stockfish the optimal solution has been to do more aggressive LMR on non-pv nodes and less aggressive LRM on pv nodes (and on root). Null move is only done in non-pv nodes.

[Mangar]

Hi,

with Spike the same, slightly better with lmr at pv nodes.

Greetings Volker