What's Your Engine's Maximum LMR Reduction?

[Daniel Anulliero]

BTW, one thing I can guarantee. You can NOT tune this stuff with 10s +0.1s type games. We've been tuning null-move parameters, and until you get to something decent (5m+5s or so) you won't get anything useful tuning-wise. Some things look bad (particularly in self-play), some things look break-even. But at decent time controls, some of this will actually start to work. But it takes a ton of testing time.

Yes much much time !
That's why I've tested at 10s+0.3...
Now I'm running a 5'+1s like hg tourney

[Henk]

BTW, one thing I can guarantee. You can NOT tune this stuff with 10s +0.1s type games. We've been tuning null-move parameters, and until you get to something decent (5m+5s or so) you won't get anything useful tuning-wise. Some things look bad (particularly in self-play), some things look break-even. But at decent time controls, some of this will actually start to work. But it takes a ton of testing time.

Yes much much time !
That's why I've tested at 10s+0.3...
Now I'm running a 5'+1s like hg tourney

Ok say 10.000 games. That will be 100.000/(60 x 24) = 70 days. We just have too wait a few months.

But then you find an error in your code or worse in your test and you can start all over again.

[bob]

BTW, one thing I can guarantee. You can NOT tune this stuff with 10s +0.1s type games. We've been tuning null-move parameters, and until you get to something decent (5m+5s or so) you won't get anything useful tuning-wise. Some things look bad (particularly in self-play), some things look break-even. But at decent time controls, some of this will actually start to work. But it takes a ton of testing time.

Yes much much time !
That's why I've tested at 10s+0.3...
Now I'm running a 5'+1s like hg tourney

Ok say 10.000 games. That will be 100.000/(60 x 24) = 70 days. We just have too wait a few months.

But then you find an error in your code or worse in your test and you can start all over again.

Or you can introduce significant regressions to longer time control games and not even know. That's far worse.

[bob]

Reduction = 0.3 * e ^(0.112*legal_move)

Strange reduction formula. How did you come up with it?

dubious...

better concave than convex IMO.

That looks concave to my eye. Did you look at the graph?

It increases slowly as number of moves goes up, but the rate of increase climbs as you get way out there.

[lucasart]

Maybe in English "concave" means "convexe" in French then.

For me, convex means f''(x) >= 0. That's not what you want for LMR. You want a function where the growth rate decreases (f''(x) <= 0).

[AlvaroBegue]

Maybe in English "concave" means "convexe" in French then.

For me, convex means f''(x) >= 0. That's not what you want for LMR. You want a function where the growth rate decreases (f''(x) <= 0).

The convention seems to be the same in English: http://en.wikipedia.org/wiki/Convex_function

[Daniel Anulliero]

BTW, one thing I can guarantee. You can NOT tune this stuff with 10s +0.1s type games. We've been tuning null-move parameters, and until you get to something decent (5m+5s or so) you won't get anything useful tuning-wise. Some things look bad (particularly in self-play), some things look break-even. But at decent time controls, some of this will actually start to work. But it takes a ton of testing time.

Yes much much time !
That's why I've tested at 10s+0.3...
Now I'm running a 5'+1s like hg tourney

Ok say 10.000 games. That will be 100.000/(60 x 24) = 70 days. We just have too wait a few months.

But then you find an error in your code or worse in your test and you can start all over again.

I don't care to play 10000
I try to play 500-600 games , It's just a hobby , don't take it too seriously
And for Mr Bob :
5 min +1 sec test running :
After (only) 25 games
The "hgm tourney " vs "bug fix " : 15-10
10" +0.2 seems too few

[Henk]

BTW, one thing I can guarantee. You can NOT tune this stuff with 10s +0.1s type games. We've been tuning null-move parameters, and until you get to something decent (5m+5s or so) you won't get anything useful tuning-wise. Some things look bad (particularly in self-play), some things look break-even. But at decent time controls, some of this will actually start to work. But it takes a ton of testing time.

Yes much much time !
That's why I've tested at 10s+0.3...
Now I'm running a 5'+1s like hg tourney

Ok say 10.000 games. That will be 100.000/(60 x 24) = 70 days. We just have too wait a few months.

But then you find an error in your code or worse in your test and you can start all over again.

I don't care to play 10000
I try to play 500-600 games , It's just a hobby , don't take it too seriously
And for Mr Bob :
5 min +1 sec test running :
After (only) 25 games
The "hgm tourney " vs "bug fix " : 15-10
10" +0.2 seems too few

Actually I meant one run is not enough for you don't know the right formula. So it will take a hopeless amount of time. Or stop if the result does not look too bad. At this moment I disabled LMR for my current move ordering is hopeless too and I even don't know right formula for null move reduction. Not to talk about a worthless killer moves implementation. So enough other things to do.

Better first be sure that other code works right before starting almost infinitely long tuning operations. How often I encountered that a test failed after running it all day long. Such a waste.

[Daniel Anulliero]

BTW, one thing I can guarantee. You can NOT tune this stuff with 10s +0.1s type games. We've been tuning null-move parameters, and until you get to something decent (5m+5s or so) you won't get anything useful tuning-wise. Some things look bad (particularly in self-play), some things look break-even. But at decent time controls, some of this will actually start to work. But it takes a ton of testing time.

Yes much much time !
That's why I've tested at 10s+0.3...
Now I'm running a 5'+1s like hg tourney

Ok say 10.000 games. That will be 100.000/(60 x 24) = 70 days. We just have too wait a few months.

But then you find an error in your code or worse in your test and you can start all over again.

I don't care to play 10000
I try to play 500-600 games , It's just a hobby , don't take it too seriously
And for Mr Bob :
5 min +1 sec test running :
After (only) 25 games
The "hgm tourney " vs "bug fix " : 15-10
10" +0.2 seems too few

Actually I meant one run is not enough for you don't know the right formula. So it will take a hopeless amount of time. Or stop if the result does not look too bad. At this moment I disabled LMR for my current move ordering is hopeless too and I even don't know right formula for null move reduction. Not to talk about a worthless killer moves implementation. So enough other things to do.

Better first be sure that other code works right before starting almost infinitely long tuning operations. How often I encountered that a test failed after running it all day long. Such a waste.

Hey ! You forgot possibles bugs in hashtables too!
Sûre there are lots of dificult things to tune in chess programming but now we're a lot off topic..

[lkaufman]

BTW, one thing I can guarantee. You can NOT tune this stuff with 10s +0.1s type games. We've been tuning null-move parameters, and until you get to something decent (5m+5s or so) you won't get anything useful tuning-wise. Some things look bad (particularly in self-play), some things look break-even. But at decent time controls, some of this will actually start to work. But it takes a ton of testing time.

Have you found that LMR reductions should be more at five minutes than at ten seconds, or less, or is there no pattern? Also, have you found that LMR reductions should be more against a gauntlet than in self-play, less, or no pattern? Same questions for null-move reductions? Finally, what is the evidence for your above statement? After all, Stockfish never tests anything at longer than one minute plus 0.05 seconds, and it's not such a weak program.
Thanks in advance for your comments.