LMR

[Bas Hamstra]

This percentage means nothing in this context. That reducing a move is ok 95% of the time means it's wrong 5% of the time. Like in a critical combination that you will now overlook. One such a mistake can be totally lethal. Like going from +2 to -6 the next move, as I see a lot lately at the server.

Bas



> It is not exactly for "no good reason". Most nodes either fail high or fail > low. If they fail high, it happens on the first move most of the time
> (>90% at least). If it fails low, every move must be searched,
> regardless of effort required. LMR actually seems quite natural when
> you think about it.

[adieguez]

Hi.

"not even a 100 game match"? I used gauntlets of >500 games.

Anyway am sorry, I was too optimistic just because it has worked for me so easily. You said after the 4th move, how good are your first 4 moves? it depends on your history or killers implementation. Fix your history, or if you use just 2 or 3 killers, make more killers. Reductions are dominating now so it doesn't matter if making more killers doesn't make your "normal" version stronger. Anyway every program is different.. ok.

But have you tried it in a long match?

Yes...it gives me nothing. Not even a 100 game match.

I also read older posts of Bob Hyat in which he tried and tried and tried. And still reported it gave nothing. I don't know what caused his position to change. Suddenly he was a defender, I don't understand.

What I don't understand is that people claim FHR improves tactical ability. In all my testsuites critical winning (but unexpected) moves are found way later in time.

I see at the server programs report ridiculous depths, presumably using LMR, and making HORRIBLE blunders. Like going from +2 to -6 the next move

>my very first and simple try with LMR gave me 34 ELO. My current >implementation (just my 4th try and still I have many things to try) is >giving me 45 ELO.

I think there may be some room for gain, but NOT the naive unrestricted implementation as propagated in this trhead. It seems yours is adapted too.

Naive version: reduce everything after the 4th move, provided the move is no capture and there was no extension triggered. It sucks

Bas

[Gerd Isenberg]

Any non-believers in LMR? It seems to introduce some nasty randomness, missing tactics here and there. Well, how could it be good, reducing moves upon moves for no good reason?
Bas

Welcome Bas,
I am on your side as well

Cheers,
Gerd

[Bas Hamstra]

&& !dangerous

Yeah, but how to define "dangerous"? That's the real question. Yes, I am working again a bit on Tao, after a pause of several years. To do what I really want I need a complete rewrite and abandon bitboards. I started with it, but its a long way to go. In the mean time I tweak the old Tao a bit

Bas

[Bas Hamstra]

Hi Gerd, you old friend! I miss the good old dutch CC championship and feel nostalgic about the atmosphere there I remember Tony (after quite a number of beers) fixing bugs in a program of a friend of mine, in the hotelbar, very late at night

Bas

[Dann Corbit]

&& !dangerous

Yeah, but how to define "dangerous"? That's the real question. Yes, I am working again a bit on Tao, after a pause of several years. To do what I really want I need a complete rewrite and abandon bitboards. I started with it, but its a long way to go. In the mean time I tweak the old Tao a bit

Bas

Why do you want to abandon bitboards? They are really coming into their own with 64 bit hardware and operating systems. Maybe you can just switch from rotated bitboards to magic bitboards or something like that.

[bob]

This percentage means nothing in this context. That reducing a move is ok 95% of the time means it's wrong 5% of the time. Like in a critical combination that you will now overlook. One such a mistake can be totally lethal. Like going from +2 to -6 the next move, as I see a lot lately at the server.

Bas



> It is not exactly for "no good reason". Most nodes either fail high or fail > low. If they fail high, it happens on the first move most of the time
> (>90% at least). If it fails low, every move must be searched,
> regardless of effort required. LMR actually seems quite natural when
> you think about it.

Look more closely. Yes you will make an occasional mistake with _any_ type of approach that trims the search tree. But there is a corresponding gain as well, in that other branches are searched even more deeply than normal, which means you will spot things now that you could not spot prior to LMR and/or forward-pruning approaches.

Perfection doesn't exist in computer chess...

[Dann Corbit]

This percentage means nothing in this context. That reducing a move is ok 95% of the time means it's wrong 5% of the time. Like in a critical combination that you will now overlook. One such a mistake can be totally lethal. Like going from +2 to -6 the next move, as I see a lot lately at the server.

Bas



> It is not exactly for "no good reason". Most nodes either fail high or fail > low. If they fail high, it happens on the first move most of the time
> (>90% at least). If it fails low, every move must be searched,
> regardless of effort required. LMR actually seems quite natural when
> you think about it.

Look more closely. Yes you will make an occasional mistake with _any_ type of approach that trims the search tree. But there is a corresponding gain as well, in that other branches are searched even more deeply than normal, which means you will spot things now that you could not spot prior to LMR and/or forward-pruning approaches.

Perfection doesn't exist in computer chess...

I think it is like null move. Will any serious computer chess program really do without it? But if you implement it blindly, it will bite you because zugzwang is not uncommon in chess endgames. So you need some caveats even in a simple test like that. Similarly with LMR or any other idea whose purpose is to trim the tree.

[bob]

But have you tried it in a long match?

Yes...it gives me nothing. Not even a 100 game match.

I also read older posts of Bob Hyat in which he tried and tried and tried. And still reported it gave nothing. I don't know what caused his position to change. Suddenly he was a defender, I don't understand.

You are probably confusing things that I wrote. I did write that "history-based LMR" did not work. However, I have repeatedly run tests and reported here that if you add LMR on top of null-move, it is worth about +40 Elo. If you don't use null-move, the gain is +80. The two are somewhat complementary in what they do so the effect of doing both is not the same as the effect of doing just one (null-move or LMR are worth about +80 by themselves).

What I don't understand is that people claim FHR improves tactical ability. In all my testsuites critical winning (but unexpected) moves are found way later in time.

I see at the server programs report ridiculous depths, presumably using LMR, and making HORRIBLE blunders. Like going from +2 to -6 the next move

I saw that in pre-null-move days, in post-null-move days, and the same for LMR. You can easily stumble into a brilliant win or an ugly loss, needing only one more iteration to see either.

>my very first and simple try with LMR gave me 34 ELO. My current >implementation (just my 4th try and still I have many things to try) is >giving me 45 ELO.

I think there may be some room for gain, but NOT the naive unrestricted implementation as propagated in this trhead. It seems yours is adapted too.

Naive version: reduce everything after the 4th move, provided the move is no capture and there was no extension triggered. It sucks

Bas

I'd never do that. I don't reduce the hash move, nor captures, nor killer moves, moves that give check, etc. Anything else is fair game, if I can't see any tactical potential for the move.

[Gerd Isenberg]

Hi Gerd, you old friend! I miss the good old dutch CC championship and feel nostalgic about the atmosphere there I remember Tony (after quite a number of beers) fixing bugs in a program of a friend of mine, in the hotelbar, very late at night

Bas

No problem to fulfill your nostalgic desires. Take the 16th off this month and register here! Would be great to meet again.

Gerd