03 March 2011

Learn from Your Engines?

Continuing with Arena, Rybka, Houdini, & Scid, I installed all of the above named software to my main machine and tailored everything to suit my taste. Since I had never used Houdini before, I decided to run it against one of the games from Learn from Your Losses?, where I listed all of my losses during my first eight years of correspondence play. I chose the last game on the list (Soreghy - Weeks, a Modern Benoni), then spent an hour stepping back & forth through the game with Houdini, trying to find some point where I had gone wrong.

The first diagram shows the game just after I departed from theory. After 13.Bc1-f4, I had found a couple of games with 13...Qc7 and another couple with 13...c4, so I played 13...Qb6 instead, not finding anything disadvantageous about the move. Houdini agrees that the position is equal.

After another 20 moves, the game reached the position shown in the second diagram. White played 34.Rc4, with a definite advantage that was eventually converted into a win. Between 13...Qb6 and 34.Rc4, Houdini offered many alternate moves, but there was no position where I had played a move that was obviously inferior.

It seems that my 'mistake' was to play reasonable moves that didn't fit into any overall plan to improve my position. To compare it to a game of tennis, it was as if I had just been content to return the ball without ever putting any real pressure on my opponent. It was my opponent who eventually put the pressure on me and I cracked.

How can I use an engine to improve on this aspect of my play? It's not at all clear to me.


Tom Chivers said...

Following the automatic computer analysis at chessbomb makes this kind of drifting clear. Sometimes every single move of a game will be coloured blue (i.e., the move is good) but the evaluation will slowly drift from equal to lost for one side.

ChessClues said...

I'm not sure how useful engines are in improving play.

From the difference in positions between move 13 and 33, i'm guessing there were a few moves white made with an exchange and gain of tempo or gain of space and tempo.

How does the computer evaluate those? Is the computer able to evaluate the cumulative effect of these?