Chess engine

Sort:
Avatar of skf95

Many chess engines can calculate multiple lines at the same time. But the engine will give stronger moves if it only evaluates one, I've heard.

Why is so? What if the engine finds a very good move and analyzes it forever, but it turns out that if the engine had gone, for instance, "depth 32" on another line, that would be better?

So my question being; When calculating 1 line, will the engine exclude all other moves than the first "very good move" it finds? If not, what's the difference between evaluating 1 and multiple lines?

Avatar of watcha

I used to believe the same thing ( engines being stronger in single PV mode than multi PV ).

But confronted with examples to the contrary I must say that there exist cases in which the correct move can only be found in multi PV mode by an engine.

I presented an example here ( see analysis of the Topalov - Shirov game ):

http://www.chess.com/forum/view/general/positions-engines-get-wrong--please-contribute

Avatar of skf95

The question is asked because of the puzzle above. The Stockfish engine (running on my Macbook Pro) can't see Rxb6 is winning (mate in 10, in fact) when calculating 1 line. If it calculates 7 lines, however, it will find the mate in 10 in a minute or so.

PS: There are different approaches after Rxb6, but they are all mate in 10. 

Avatar of watcha

This is a common phenomenon when a sacrifice is the correct move. Since a sacrifice loses material in the short term such moves will be valued much lower at small depth than 'normal' moves. The engine uses the shallow search results for move ordering so the move is not examined further.

If you force the engine to make the sacrifice move ( by raising the multi PV level to the point where the sac move becomes viable ) it may immediately find the correct line.

In the thread I mentioned the correct move for the Topalov - Shirov game is only found at multi PV 14 at depth 24.

Avatar of skf95

So in chess engine tournament, one has to decide how many lines the engine should evaluate? In most sitataion I guess PV <4 or so is the best, but in some cases, PV >10 is neccessary!

Avatar of Fugazy_Crapov
skf95 wrote:
 

The question is asked because of the puzzle above. The Stockfish engine (running on my Macbook Pro) can't see Rxb6 is winning (mate in 10, in fact) when calculating 1 line. If it calculates 7 lines, however, it will find the mate in 10 in a minute or so.

PS: There are different approaches after Rxb6, but they are all mate in 10. 

An interesting position!  On my modest laptop, Houdini 3 solves the mate in under 10 seconds using the default engine settings.  

But Komodo 6 is too fixated on capturing the bishop on e1 to see the rook sac and subsequent mate!