Wanted: A successor to Chest

[Marek Soszynski]

Dear Programmers,

We have plenty of analysis engines already, most of them "free". Here is something we are genuinely short of: a dedicated mate-solver that is cross-platform, UCI, 64-bit and multi-processor. In other words, a successor to Chest; in fact a new engine altogether.

Maybe, just maybe, someone here will quietly take up the challenge of my suggestion.

(I've raised the topic before...

http://www.talkchess.com/forum/viewtopi ... w=&start=0
http://www.talkchess.com/forum/viewtopi ... ht=#257894)

[hgm]

Better if it would be WB. Who needs UCI?

[hMx]

Dear Programmers,

We have plenty of analysis engines already, most of them "free". Here is something we are genuinely short of: a dedicated mate-solver that is cross-platform, UCI, 64-bit and multi-processor. In other words, a successor to Chest; in fact a new engine altogether.

Maybe, just maybe, someone here will quietly take up the challenge of my suggestion.

(I've raised the topic before...

http://www.talkchess.com/forum/viewtopi ... w=&start=0

Dann Corbitt's direct answer hits the nail quite well: the main problem is to motivate someone to do it. It would be a lot of work, and I am not motivated, anymore. The current situation with clones of clones everywhere does not exactly help, either

Technically, the challange is "multi-processor", IMHO.

http://www.talkchess.com/forum/viewtopi ... ht=#257894)

[Jouni]

There is already one: Critter! In my test with 200 random longmovers it's clearly better than Chest.

Critter 1.6 126
Chest/XKP 92
Chest/KPX 87

But version 1.4 was better - there seems to be a minor bug now

[lucasart]

There is already one: Critter! In my test with 200 random longmovers it's clearly better than Chest.

Critter 1.6 126
Chest/XKP 92
Chest/KPX 87

But version 1.4 was better - there seems to be a minor bug now

I second that. Critter is extremely strong on tactical testsuites.

[Marek Soszynski]

There is already one: Critter! In my test with 200 random longmovers it's clearly better than Chest.

Critter 1.6 126
Chest/XKP 92
Chest/KPX 87

But version 1.4 was better - there seems to be a minor bug now :(

"Clearly better" in what respects? Remember that the point of a mate-solver is complete accuracy rather than speed. Here's what I said before:

As soon as an ordinary engine finds a single forced mate, it is "job done" - some engines don't even bother analysing any further. But because of the way engines search, the first mate found might not be the shortest. A dedicated mate solver has to look at everything (so is very slow in comparison) but will find the genuinely shortest mate, and also find out whether it is a unique solution. I smile when I see posters boasting that one engine or another (without tablebases) sees a mate in 24, say. Mate announcements by ordinary engines are not to be entirely trusted.

[Jouni]

Is it possible to add search_only_mate mode to Stockfish or any top engine? Should be faster, when no positional evaluation is needed. Houdini has special mate mode, but it's not that kind.

[Evert]

I think you're operating under a false assumption, namely that programmers of chess engines write chess engines because they like to offer something to other people. Speaking for myself, and I guess this goes for many others, this is not the case: I write a chess program for my own entertainment, and I release it for free on the off-chance that someone else may find something interesting in it.

Personally I have no real interest in writing a mate-only solver. Maybe a special mode to get Jazz to find the fastest mate if it already found a mate, but I think I'll only do that if I'm bored.

[diep]

Dear Programmers,

We have plenty of analysis engines already, most of them "free". Here is something we are genuinely short of: a dedicated mate-solver that is cross-platform, UCI, 64-bit and multi-processor. In other words, a successor to Chest; in fact a new engine altogether.

Maybe, just maybe, someone here will quietly take up the challenge of my suggestion.

(I've raised the topic before...

http://www.talkchess.com/forum/viewtopi ... w=&start=0
http://www.talkchess.com/forum/viewtopi ... ht=#257894)

With todays huge nps-es it would be possible to really do damage nowadays with advanced and special type algorithms for this; the bigger the nps the more impact clever algorithms have.

Developing them takes time though.

How much are you prepared to pay for that?

[diep]

Is it possible to add search_only_mate mode to Stockfish or any top engine? Should be faster, when no positional evaluation is needed. Houdini has special mate mode, but it's not that kind.

special algorithms total annihilate any engine there when it's just about mate.

what you can build now is something that's having effective search space speed of around a 60 million nps single socket or so.

So just nps wise you smack everything already and that's just the start, then special algorithms will do the rest.

It's not easy to develop special algorithm for this that is total superior over anything from the past. Also one would want several types of variations to also avoid zugzwangs from blocking your search and things like that.

Unforced mate in 40 with near to no checks should not be problem for this.
Probably you also would want integrate EGTBs into this.