Engine's Feature

[swami]

I voted Personality Creator.

However, I'd like better "Showing PV lines and score while the engine is thinking", for Thinker.

This not only applies to Thinker, There are other engines with more than 2700 elo but come with one or two of these features. Engines such as Bugchess, Colossus, Slowchess. Twisted Logic etc

a PV and score would be a good idea, but a lot of engines exhibit this in their analysis lines.

I also chose the personality creator, it would be good to see another engine come up with parameters just like how it was in chessmaster. So far Glaurung seems close to having the personality options but it doesn't seem to be popular in that area.

[Nimzovik]

I would opt for 1) accurate node and hash use meter, 2) a meter to indicate when hash is full -all ram used - 3) A personality on the order of the "turk" (see fritz 11) however done in Nimzovich or capabalnca etc. style that talks. 4) More eye candy in terms of sets and background. 5) The ability to make your own CD 's -with video capture-in terms of openings etc. 6) The ability to accurately assess closed positions- which after all IS chess despite some people's opinions. 7) The ability to merge any and all books created by using other programs such as obk to ctg. 8) one that administers electric shck to cheaters on the live server 9) and also negotiates paying my bills on line. Nuff said?

[Mike S.]

I voted for MP, learning, and "other" because I wasn't sure if multivariation analysis mode, aka multi-pv, is considered an extra feature. For me it is the most important feature for analysis. I think this is also engine protocol related because AFAIK the WB. protocol doesn't allow multi-pv(?). (Some Crafty versions converted to cb-native protocol, have the multi-pv feature though.)

Also, I find UCI and/or cb-native engines generally more convenient to use. I am in a process of de-nerdification.

Other points not mentioned are use of Nalimovs or bitbases, and hashlearning (not via learnfile), for better interactive "backwards" analysis. These are also important.

[Ovyron]

I voted for MP, learning, and "other" because I wasn't sure if multivariation analysis mode, aka multi-pv, is considered an extra feature. For me it is the most important feature for analysis.

Hey, could you tell us a bit about the analysis method that you're using? I've found multi-pv extremely helpful in engines like Zappa or Toga, that don't support exclude moves, so after refuting a variation, the only way to continue the analysis is to use multi-pv to ask the engine for a second best move.

However, for engines that do support exclude moves, ignoring the refuted moves seems to be much more efficient, so I am curious (I'd put "move excluding" at the top of my wishlist of extra features over multi-pv.)

[peter]

I've found multi-pv extremely helpful in engines like Zappa or Toga, that don't support exclude moves, so after refuting a variation, the only way to continue the analysis is to use multi-pv to ask the engine for a second best move.

However, for engines that do support exclude moves, ignoring the refuted moves seems to be much more efficient, so I am curious (I'd put "move excluding" at the top of my wishlist of extra features over multi-pv.)

Hi!
Sorry, there I don't get your point. If you know, which move to exclude, what do you need the engine for? Just the other way round is mine as for analyzing positions: first have a look at the score of some candidate moves, then try them on the board, then going back to see how the score changes.
To exclude all moves, you're not interested in, simply choose any you consider to be better.
There are engines that seem to be capable of calculating only one or two variants at a time, these are not my favorites for analyzing positions I don't have a plan of my own in favour. (Deep) Fritz is my best example for real multivariant mode of calculating. It's almost not slowed down showing 10 variants at a time and trying it in tricky positions in that way I'm surprised again and again how much quicker it sees solutions thus then in single variant mode.
Rybka (3) is an exception once more, as one-tracked as she sometimes seems, in tactical positon I'd call her double-tracked. Try to solve best-move-probs in 2variant mode with her. Often enough she's much faster finding hidden moves that way then in single mode and even faster then with "find win".
So my personal wish for engines or bad enough rather for GUIs was a special triple_or_more_brain_feature for automatic (backward) analysis (the one Shredder has is something completely different as a pity even if it's rather interesting too or at least would be, if different engines would really share their hashtables that way ) or even for engine-engine match. Giving an engine really capable of this enough on-time would bring up much more tactical positions against other engines probably
regards
Peter.

[swami]

I voted for MP, learning, and "other" because I wasn't sure if multivariation analysis mode, aka multi-pv, is considered an extra feature. For me it is the most important feature for analysis. I think this is also engine protocol related because AFAIK the WB. protocol doesn't allow multi-pv(?). (Some Crafty versions converted to cb-native protocol, have the multi-pv feature though.)

Also, I find UCI and/or cb-native engines generally more convenient to use. I am in a process of de-nerdification.

Other points not mentioned are use of Nalimovs or bitbases, and hashlearning (not via learnfile), for better interactive "backwards" analysis. These are also important.

You have provided interesting stuffs in there!

I should have asked you for the list of options before making this poll.

[Ovyron]

Hi!
Sorry, there I don't get your point. If you know, which move to exclude, what do you need the engine for?

The basic idea goes like this (example):

You let the engine analyze for some time the position you're interested in. It comes out with move A, as white, with a score of 0.20.

Since my time is limited, I get better results by forcing this move and see what the engine thinks about it, basically pruning all the other moves to get deeper.

It comes with move Aa, as black with a score of 0.10.

Extending the line a bit, the engine thinks this move is not as good as it seemed, and I repeat step 1. Now move AaA has a score of 0.00.

The best move the engine has found leads to 0.00 when I go deeper, so what I do, is going back to the original position, and exclude move A, so the engine searches for an alternative.

Now, the engine will come with move B, and I may choose to look it further (if its score is > 0.00) or go back to move A (if B's is score < 0.00)

This is usually faster than sitting at the original position 2 extra plies (or more) to wait for the engine to realize A is 0.00, faster than using Multi-PV 2 on each position (because in the example, we don't need to search for alternatives for black) and it's impossible to do with engines that don't support Multi-PV or exclude moves (because once the hash is full it'll start running in circles.)

[peter]

Hello!
Now I see, what you mean, but in the GUIs I use, engines that don't support multivariant mode don't react to exclude move neither, for example crafty with winboard2Uci in chessbase or Shredder doesn't. Do you know if Crafty works with exclude move in Arena?
Best regards
Peter.

[Ovyron]

No, exclude moves is a UCI feature (that the engine needs to support by itself, I.E. Zappa Mexico II it's UCI and doesn't support it) though perhaps it could be implemented in Winboard engines in the future.