UCI engines with fixed node count support!

[Edsel Apostol]

Hi everyone,

I'm planning to try engine testing with fixed nodes. I'm wondering what are the UCI engines that support this. I only know about Glaurung and I have not even tried if it's working.

I didn't put it in the readme.txt included with the Twisted Logic package but it supports this UCI feature including "go searchmoves". That means if the GUI supports it, the engine could limit only its search to the moves the user wants in the root.

To the moderators: I don't know if this thread is okay here in the General Topics. Feel free to move it in Programming or in the Tournament sub-forum.

[Marc MP]

[quote="Edsel Apostol"]Hi everyone,
...

I didn't put it in the readme.txt included with the Twisted Logic package but it supports this UCI feature including "go searchmoves". That means if the GUI supports it, the engine could limit only its search to the moves the user wants in the root.

To the moderators: I don't know if this thread is okay here in the General Topics. Feel free to move it in Programming or in the Tournament sub-forum.quote]

Hi Edsel,

"go searchmoves" is a function I used very often with Chess Tiger 2007 and Glaurung 2.1. It saves tons of time in many positions. I'm very likely to try latest releases of Twisted Logic more often for position analysis now. Thank you for the information.

[Edsel Apostol]

Hi Marc,

Tell me if you have problems with that feature. It has been there for a long time but it hasn't been well tested.

[Bill Rogers]

I am not aware of any engine that allows fixed nodes, but most will allow a fixed depth with the exception of their quesicent extensions. These are my main interest as they tend to point out an engine true strength, at least in my opinion.
Bill

[Uri Blass]

Hi everyone,

I'm planning to try engine testing with fixed nodes. I'm wondering what are the UCI engines that support this. I only know about Glaurung and I have not even tried if it's working.

I didn't put it in the readme.txt included with the Twisted Logic package but it supports this UCI feature including "go searchmoves". That means if the GUI supports it, the engine could limit only its search to the moves the user wants in the root.

To the moderators: I don't know if this thread is okay here in the General Topics. Feel free to move it in Programming or in the Tournament sub-forum.

Movei supports fixed nodes as uci engine.

Uri

[Edsel Apostol]

Thanks Uri. I will use Movei and Glaurung then as sparring partner for Twisted in this testing scheme.

[hgm]

I am not aware of any engine that allows fixed nodes, but most will allow a fixed depth with the exception of their quesicent extensions. These are my main interest as they tend to point out an engine true strength, at least in my opinion.
Bill

If there was a prize for wildly untrue statements, this one would qualify!

Someone was testing Joker80 against Gothic Vortex for me, and he insisted doing this at a fixed depth (which initially he set to 12 ply). The result was that Vortex was allowed to to use about 10 times as much time per game as Joker80 (and,predictably, of course won most games by outsearching Joker by 2 or 3 ply).

If you set an engine for 12 ply, you can be almost certain about one thing: virtually no line will be searched to 12 ply. Vortex was searching most lines 12-13 ply, because it apparently uses many extensions. (And if some extensions were triggered close to the root, e.g. because the PV started with two checks, sometimes took a hundred times longer to complete the search for that move.) While Joker80 was searching most lines to only 10 ply (because it has virtually no extensions, and uses Late-Move Reductions on almost all moves).

As a consequence, the whole exercise did not tell us anything about the relative strength of the engines under ordinary tournament conditions, except that it is a disadvantage to get almost no time. We are repeating now under conditions where the depth of Joker80 is set to 14 ply, and that of Vortex to 12. Then they use about equal time measured over the entire game.

[Marc MP]

Hi Marc,

Tell me if you have problems with that feature. It has been there for a long time but it hasn't been well tested.

Hi Edsel,

Unfortunately it doesn't seems to work properly in ChessPartner. I used the following position:

[d]r2r4/qb3k1p/p3pn2/n3b1Q1/Pp1PP3/8/1B1NBPPP/3R2K1 w - - 2 1

I did: Extra -> Analyze position -> Include Moves -> I thicked Bh5+ and dxe5 -> OK.

Prior to Depth=14, it seems to work: I can see in the GUI that the move analyzed is either dxe5 or Bh5+. But then from Depth=14 onward, twisted goes through the full move list at each iteration.

[Edsel Apostol]

Hi Marc,

It would be great if you could send me the logfile of Twisted after you run that position, or maybe post it here. I don't have a GUI here that supports that so I couldn't test.

I initialize the movelist by the moves received through the "go searchmoves" if the engine received that command so if it worked in the first 13 plies that means it should work in plies above that as the movelist is only initialized in the beginning and it should not change.

I will try to review my code, maybe I overlook something.

[Bill Rogers]

Mr. Muller
As I had mentioned some programs will search deeper because of their quiensence searches. I regularly use TSCP and Minimax, to very simple program that are not rated very high as I test a lot of really weak engines.
On both of those programs during a Six Ply setting I have noticed that both will from time to time search as deep as 10 plys do to their quiensence extentins. Most of the time they only go 1 or 2 levels deeper, maybe 3 depending on how active or quiet the position might be.
As to why Joker when ordered or commanded to do a 12 ply search and then doesn't, I can't begin to answer that, in fact this is the first time that I have ever heard of a program ignoring a set perimiter like that. In you case you must be right but as I said I never heard of such a thing before now.
Bill