Thursday, May 19, 2011

Engine Analysis Methods

Most people running an engine analysis of a game simply have Fritz run a full analysis for a set time limit per move or a set ply-depth, but that does not always give satisfactory results.  Below is a list of some methods that will give better results, but in the end, if one is going to learn anything or hope for decent results in high level CC events where engine use is allowed, then it is going to require interactive analysis where you work with the engine in trying to ferret out the best move in any position. And that is going to require more than just plugging a game into Fritz and letting it do its thing. It’s going to require patience and, as you will see, making some judgment calls based on your understanding.  Anyway, I hope this brief guide will be of use in assisting readers in understanding how to analyze a game with an engine and perhaps help in using engines in their quest for improvement.


Sparring Partner
      You come up with a plan and use the engine as a blundercheck to see if the move loses material. This is a good way to check out your ideas because engines will put up tough resistance to you moves. This method is also effective because sometimes we get so carried away with our ideas that we can overlook the best moves but the engine will not.
      Another way of using this interactive analysis is, once you have a line you want to analyze, go to the end and keep going back one move and wait to see if the engine’s analysis makes a large change. If it does then you will want to check out those side lines to make sure the engine did not miss anything in its original analysis.
      Note: if the engine pauses for a long time and/or it keeps jumping back and forth between candidate moves, it means it is having a hard time deciding on the move it thinks best. For you, this means it’s a good place to stop and investigate further the move the engine was considering.


Multivariations
      In this mode you click on the + or – to increase/decrease the number of moves variations displayed. When the difference between the first and second choice is about one Pawn and the position is relatively quiet tactically, then the conclusion is that the first choice is probably best. However, in a tactically wild position before concluding that the first choice is clearly best, you need a difference of 2 or 3 Pawns before concluding that the first choice is definitely the best one. If the difference is about a half Pawn, you can’t draw any definite conclusions.


Multiple Engines
      Running more than one engine at the same time is one of the best ways to analyze a position because different engines will not always come up with the same first choice. When you get different suggestions, you will have to investigate them further and make a decision as to which move offers the best possibilities. Another point to understand is that when it comes to positional evaluations, engines can vary greatly in the numbers they display. Usually the difference will only be a half-P or less which is insignificant, but it can also be quite large. When this happens we usually either go with the engine showing the largest difference or the one that’s rated the strongest, but this can be misleading in positions where there are material imbalances but one side has attacking possibilities. So the question then becomes one of which engine are you going to believe?


Engine Tournaments
      Running a tournament with multiple engines from a given position is another way of testing a position. After the tournament you can look at the games to get an idea of different liens of play and see where they lead. One word of caution though…the results can be skewed by later engine errors, so don’t take the results at face value without checking them out first. The most effective positions to evaluate in this manner are those in which the next few moves will affect the game in some way. Examples: which side to castle on, P-moves that close the center or result in isolated or doubled P’s, exchanges leading to material imbalances, or positions that are complicated and open.


Deep Position Analysis
      This method of analysis is one that establishes an analysis tree and you can adjust the length of the branches. This method is probably the least desirable method and is best used when you don’t want to spend time interacting with the engine and are content to just let the engine do all the work. On the other hand if one’s understanding of chess isn’t very good this method is satisfactory.

Full Analysis and Blunderchecking
      This is good if you want to check a game for tactical errors, but remember that this analysis will be pretty meaningless in closed positions and setting blunderchecking at any value of a half-P or less will result in “improvements” that are, generally, meaningless.

1 comment:

  1. Great post!

    Whenever I am analyzing one of my games, or a master game I always go through the game two or three times on MY OWN and then I check the main or questionable points with a couple of engines. And then if the engines disagree I work through their variations on my own to see which one makes the most sense to me.

    I also like to to find a position in a master game and write down everything I can think about it, come up with a plan and then play it out against an engine.

    Thanks again for the great post!

    ReplyDelete