LMR at root of search tree - worthwhile?

[silentshark]

The subject says it all!

Not tried this, is it worth a punt?

[hgm]

I never was able to measure any benefit from LMR in the first place. I always have done it in the root. It never occurred to me to treat the root differently from any other node.

[zamar]

Stockfish gained ~15 elo we started to use LMR at the root.

[Mincho Georgiev]

Opposite, for me LMR always worked very well but never at the root. I remember trying it at the root with absolutely no success. It's not necessary however to fail for you as well. Give it a try and post the results, would you.
P.S. if the gain is 15 elo, as Joona said, probably I wasn't able to measure it since I trow it out to soon. Who knows.

[Aaron Becker]

In Daydreamer it gave me a statistically insignificant positive result (<10 elo). I kept it in because I like to treat root nodes similarly to interior nodes unless there is a compelling reason not to.

[Daniel Shawul]

I got small improvement from it.
Caution : Root move ordering could be messed up if you use nodes count to sort the moves. I had this bug http://talkchess.com/forum/viewtopic.ph ... ht=#334554
Sorting on the cummulative nodes count from all the previous iterations (instead of just the previous one) solved this problem.

[bob]

The subject says it all!

Not tried this, is it worth a punt?

Been using it at _all_ nodes in the tree since I started. It has a bigger impact at the root since the resulting sub-trees are all very deep. Not sure why you would want to do it at 2 and up but not at 1. There are lousy moves at the root that can safely be reduced.

[bob]

I got small improvement from it.
Caution : Root move ordering could be messed up if you use nodes count to sort the moves. I had this bug http://talkchess.com/forum/viewtopic.ph ... ht=#334554
Sorting on the cummulative nodes count from all the previous iterations (instead of just the previous one) solved this problem.

Note that the "L" in LMR means "late". I would likely not want to reduce the second move I search, which could be the previous best move.

[hgm]

I should add that in my Xiangqi engine LMR improved it enormously. (Also done at the root.)

[silentshark]

cheers, guys, that sounds like it's worth a try. I will give it a go and post my results. I'm still ordering by #nodes@previous ply at the root, but don't see why I can't use LMR there too.

Interesting to see the +15 elo figure for stockfish..