Identifying combinations in a game

[jwes]

Not necessarily. Last move by the opponent could be (and quite likely is) a blunder. But okay, then one could argue that those obvious moves are also part of combiation, hmm...

I don't claim this alghortim is ALREADY 100% correct, but it's a first cut. If someone implements it and checks some games I think should be not difficult, with the REAL RESULTS in our hands to tweak the last bits and have a workeable chess combination detector.

The real problem is that the more interesting a combination is, the harder it is to find. How long do you search each move before you decide there is no combination? Another problem is endgames with deep promotions may look like combinations.

[Tord Romstad]

Defining what is and isn't a combination is a bit slippery and what you wind up with may not be so practical to efficiently use in an algorithm.

You guys should get a copy of Yury Averbakh's classic book Chess Tactics for Advanced Players. I no longer have the book, but I remember that he spent a lot of pages to give a formal definition of a combination, justifying his definition, and classifying different types of combinations. There are probably better books for teaching pratical tactics to human players, but I have never seen anything nearly as precise from a formal point of view.

Tord

[mcostalba]

The real problem is that the more interesting a combination is, the harder it is to find. How long do you search each move before you decide there is no combination?

Decopuling of alghoritm itself from alghoritm parameters is a key here, it was not a casual choice.

You can have a workable alghoritm and THEN tuning the parameters to something sound for you.

I would think that without a test run on real games it is very difficult to continue to speculate on it, but from a "having fun discussing" point of view (that is absolutely acceptable BTW).