Sunday, September 27, 2015

Hou what?

An occasionally entertaining feature of Chessdom's World Cup coverage is the three computer programs which sometimes offer quite different assessments.

I particularly liked this moment from the first Mamedyarov-Karjakin game, Komodo and Stockfish both substantially favouring White: it's not just that the evaluations are different, but that they're basically opposite, with Houdini supporting Black by a roughly similar margin. I would love to know why...if I thought there was any chance I'd understand.


Anonymous said...

With three engines and four moves each, that's twelve different ideas. Maybe the engines agree that a move is in the top four but disagree on its assessment. The screen shot is suggesting a depth of 14 ply, meaning, I think, seven moves by each player.

It's showing that engines use different factors to assess positions. If they can find something decisive, they are all likely to agree. Equally if the position is so boring that any one of half a dozen moves will do, they are likely to agree. It's where the position is double edged, that the weighting of different factors will cause divergence.


ejh said...

Yeah I kind of know that Rog but what puzzled me here was the nature of that divergence - one engine strongly contradicting the evaluation of the other two.

gambit said...

processing .time -5.sek -to not enough to properly assess the movement .Engine to correctly assess needs a few minutes.