Is Singular Extensions a special case of LMR ?

[Henk]

If I translate LMR into early move extensions then the answer will be " Only if singular moves are always early moves" and that is normally not true unless you have a move ordering that puts singular moves in early move positions.

[lucasart]

A last other question: Do you know what the tactical connected quasi Botwinnik-Markov extension approximately might add to the program strength measured in ELO if well implemented.

I tried it out a month ago, removed it for I did not notice any difference but that says nothing for you know it had been badly tested.

Botvinnik-Markov extension is plain stupid. I have never seen any evidence of a program gaining elo from that technique. In fact, in its original form, it is a very measurable regression. My advice is: don't waste your time with BM-extension, it simply does not work. It's one of those "great ideas on paper", so people like to write theoretical papers on that, but it simply does not work in the real world (as opposed to the academic world).

What you're calling "tactical connected quasi Botwinnik-Markov extension" should really be called the Tord Romstad extension, because it's an original idea of Tord, and I don't see any reason to credit Botvinnik or Markov for it. It is still in use, in its original form, in Stockfish today (unchanged since Glaurung). But, however smart and aesthetically pleasing it may seem, even this less costly version is worth 0 ELO in Stockfish: proven by serious testing.

On the other hand, removing it is not a clear gain either, so Marco decided to keep it, to be on the safe side. Personally, I would have removed it without hesitation, but Marco is paranoid about the risk of regression, so he prefers to not remove anything even useless, just in case it helps at some long TC far beyond what we can test.

[lucasart]

After some long and unfruitful experimentation, I finally found a version of using the "null search fail low" information in DiscoCheck that was an elo gain (although a tiny one, barely measurable). It's like a mate threat extension, without doing the zero window mate threat search after the null move. The idea is to rely on fail soft scores: theoretically heretic, but in practice detects some (not all) mates at low depth, and in a way that is completely free of overhead, which is why it ends up being positive.

[Tom Likens]

A last other question: Do you know what the tactical connected quasi Botwinnik-Markov extension approximately might add to the program strength measured in ELO if well implemented.

I tried it out a month ago, removed it for I did not notice any difference but that says nothing for you know it had been badly tested.

Botvinnik-Markov extension is plain stupid. I have never seen any evidence of a program gaining elo from that technique. In fact, in its original form, it is a very measurable regression. My advice is: don't waste your time with BM-extension, it simply does not work. It's one of those "great ideas on paper", so people like to write theoretical papers on that, but it simply does not work in the real world (as opposed to the academic world).

What you're calling "tactical connected quasi Botwinnik-Markov extension" should really be called the Tord Romstad extension, because it's an original idea of Tord, and I don't see any reason to credit Botvinnik or Markov for it. It is still in use, in its original form, in Stockfish today (unchanged since Glaurung). But, however smart and aesthetically pleasing it may seem, even this less costly version is worth 0 ELO in Stockfish: proven by serious testing.

On the other hand, removing it is not a clear gain either, so Marco decided to keep it, to be on the safe side. Personally, I would have removed it without hesitation, but Marco is paranoid about the risk of regression, so he prefers to not remove anything even useless, just in case it helps at some long TC far beyond what we can test.

No one has more respect for Tord than I do, but the idea orginated with Sergei Markoff, the creator of SmarThink. Tord and a few others ran with the idea, but Sergei came up with it and deserves credit for it, hence the name. If I remember right he was inspired by Botvinnik's book "Computers, Chess and Long-Range Planning". And this was long before Talkchess ever existed, back when it was CCC with a primitive, mainly text interface.

Of course Tord popularized it, but you're right there was a lot of ambiguity on how effective it was.

regards,
--tom

[Uri Blass]

A last other question: Do you know what the tactical connected quasi Botwinnik-Markov extension approximately might add to the program strength measured in ELO if well implemented.

I tried it out a month ago, removed it for I did not notice any difference but that says nothing for you know it had been badly tested.

Botvinnik-Markov extension is plain stupid. I have never seen any evidence of a program gaining elo from that technique.

It seems that old Gothmog of Tord from 2003 gained some elo from that technique.
It does not mean that botvinik-Markov extension helps modern programs but I do not see a reason to call it plain stupid and the only thing that you can say is that it does not help stockfish based on testing.

http://www.stmintz.com/ccc/index.php?id=319172

[lucasart]

No one has more respect for Tord than I do, but the idea orginated with Sergei Markoff, the creator of SmarThink. Tord and a few others ran with the idea, but Sergei came up with it and deserves credit for it, hence the name. If I remember right he was inspired by Botvinnik's book "Computers, Chess and Long-Range Planning". And this was long before Talkchess ever existed, back when it was CCC with a primitive, mainly text interface.

Of course Tord popularized it, but you're right there was a lot of ambiguity on how effective it was.

regards,
--tom

Yes, indeed. What I was trying to say is that Botvinnik has not invented anything. He wrote a chess book, in which he explained that refuting what the opponent is threatening is generally a good idea. But every chess player knows that. It's a trivial and vague assertion, hardly an "invention". There's even a pedantic name for it: prophylactic play. With all due respect, Botvinnik did not invent anything.

The one who should get credit for figuring out how to adapt this general and vague chess principle into a chess algorithm is indeed Sergei Markoff. And the one who should get credited for making the Markov extension actually work (or "kind of work" your mileage may vary) is Tord.

The devil is in the detail, and it's a far greater achievement to understand where the tradeoff lies, and find the right version of the idea than to make general statements about chess playing. That's why Tord's contribution is much higher (to my eyes) than Botvinnik's here, so I don't see why Botvinnik's name should be there instead of Tord's.

IOW: Markov-Romstad extension is the right name for it.

[mcostalba]

Marco is paranoid about the risk of regression, so he prefers to not remove anything even useless, just in case it helps at some long TC far beyond what we can test.

Yes I am paranoid on regressions, but I am also paranoid on simplifications and removing code....it is a ongoing battle of paranoias

I have applied many simplification patches. In this case we can retest even this one at fixed games long TC and see if, today, the simplification paranoia wins above the regression one.

[mar]

Botvinnik-Markov extension is plain stupid. I have never seen any evidence of a program gaining elo from that technique.

I tried this extension some time ago and couldn't measure anything, no regression but no improvement either. So I removed it.
Btw. I've noticed that the way they describe it in stminz archive is wrong.

[hgm]

IOW: Markov-Romstad extension is the right name for it.

Well, if Seirgei Markoff inventend the, 'Markoff-Romstad extensions' might perhaps be better.

[mar]

IOW: Markov-Romstad extension is the right name for it.

Well, if Seirgei Markoff inventend the, 'Markoff-Romstad extensions' might perhaps be better.

Actually it was Tord who named it Botvinnik-Markoff