Is LMR Sound.

[Henk]

Tried LMR for the tenth time. Used a simple implementation: apply LMR when move is not a capturing move. Did not notice much improvement.
Maybe the program is more tactical, but I also saw very bad strategical lines.For instance bishop got locked in totally by its own pawns.

I cannot measure ELO. But if there is any improvement it certainly is not spectacular. I remove the LMR from the program. I don't want extra code in my program if it doesn't help much.

[Daniel Shawul]

Tried LMR for the tenth time. Used a simple implementation: apply LMR when move is not a capturing move. Did not notice much improvement.
Maybe the program is more tactical, but I also saw very bad strategical lines.For instance bishop got locked in totally by its own pawns.

I cannot measure ELO. But if there is any improvement it certainly is not spectacular. I remove the LMR from the program. I don't want extra code in my program if it doesn't help much.

Wow you tried 10 times to make it work, i am impressed really. Also even though you do not measure elo, you can tell certainly tell the improvement is not spectacular because it got a bishop locked. I missed that. Given all that indeed it may be better to remove the extra 00100b lines of code of LMR. Thank you for being such a revelation to the chess community.

[hgm]

You should not expect search to solve strategic ignorance of your engine. If it locks in its Bishop, it is because your evaluation sucks (e.g. no mobility term). No amount of search could help that, and using an efficient search rather than a crummy one would thus not prevent it in any way.

If you cannot measure Elo, nothing what you say about whether it works or not carries any weight...

[Henk]

If you cannot measure Elo, nothing what you say about whether it works or not carries any weight...

Yes you're right.

[Henk]

Daniel Shawul,

We both seem to understand why this community has such a limited number of active members.

I also do not see Vincent Diepeveen anymore. What happened to him.

[Evert]

Used a simple implementation: apply LMR when move is not a capturing move.

It's wrong. You should exclude the hash move and the killer moves from LMR.

Maybe the program is more tactical, but I also saw very bad strategical lines.

Think of it this way: tactics = search, strategy = evaluation. It's not quite that simple, but it helps get the idea of what part of your program can be expected to handle what aspect of the game.
(With a deep enough search all moves become tactical in a sense).

For instance bishop got locked in totally by its own pawns.

Unless the search is deep enough to see material loss by having a bishop that's locked out of the game, the search will not seek out positions where the bishop is not trapped. Unless the evaluation gives a penalty for the trapped bishop.
Assessing piece placement is the task of the evaluation function, not the search.

I cannot measure ELO.

Yeah you can: http://remi.coulom.free.fr/Bayesian-Elo/
You don't have to though: play games and see which version wins the most. You don't really need to know how much stronger it is, just that it's stronger.
When I first started developing Jazz, my testing condition was to play 100 games against micro-Max. No, that's nowhere near enough games for serious testing, but it seemed plenty to me at the time, certainly compared to the number of games I used to play as a human. Of course, Jazz was poor enough that I could actually look at the games, spot the mistakes, fix them and see the improvement in the next game. Can't do that these days...

But if there is any improvement it certainly is not spectacular.

Define spectacular. If I have a 5 elo improvement I'm glad, if I have a 10 elo change I'm very happy, if I get 30 elo I'm ecstatic and if I ever have a 100 elo improvement the roof comes off.

I don't want extra code in my program if it doesn't help much.

If it doesn't do anything, don't put it in.
If it helps (even a little) it seems silly not to keep it.

Of course you do need to actually measure if it's an improvement or not.

[Henk]

My evaluation does have a mobility term. Could be that there are other errors. Have to find out myself by code inspection.

[Daniel Shawul]

Daniel Shawul,

We both seem to understand why this community has such a limited number of active members.

I also do not see Vincent Diepeveen anymore. What happened to him.

Well there is your proof for those who still need it.

[hgm]

I also do not see Vincent Diepeveen anymore. What happened to him.

I met Vincent in Leiden Saturday, at ICT 13. He is working on robots, nowadays.

[Don]

Tried LMR for the tenth time. Used a simple implementation: apply LMR when move is not a capturing move. Did not notice much improvement.
Maybe the program is more tactical, but I also saw very bad strategical lines.For instance bishop got locked in totally by its own pawns.

I cannot measure ELO. But if there is any improvement it certainly is not spectacular. I remove the LMR from the program. I don't want extra code in my program if it doesn't help much.

LMR is worth over 100 ELO in Komodo and is the biggest breakthrough in computer chess in years.

However, there are some things you could be doing incorrectly. Do you do the researches with the proper windows? How certain are you that the code is correct?

You should probably try something more conservative than what you are doing. First of all, you should NEVER reduce the first move. You said everything but captures, but you should probably start with an implementation that does not reduce the first several moves, perhaps 5 or 6.

Also don't reduce captures, checks or out of checks. If everything is correct a conservative implementation should definitely help your program.

What do you mean by not be able to measure ELO? You are probably getting some ELO if you are doing this reasonably well, even if it's only 20 or 30 ELO. but you won't see that unless you can "measure" it in your program. You need to play at least a few hundred games.