Engine analysis

Home

Evaluating Chess Engine Analysis

Written in 2000

Many of you that read the reviews on my website have probably used a computer to analyze your games at one time or another. We all know that they can be invaluable tools, especially when you don’t have access to a strong human player. However, while today’s chess engines are incredibly strong, they still have some weaknesses and tendencies that you need to be aware of.

This article attempts to discuss some of the things that I have observed over the years when analyzing my own games. I have literally done hundreds of post-mortems on my losses. In the course of doing this, some patterns have emerged, and I think that the readers of this page could probably benefit from my observations.

Don’t let this article give you the wrong impression. Chess engines are extremely strong, and it takes a much better player than I to beat one. But, if you use a chess engine for the analysis of your own games, you might want to keep the following information in mind – it may help you to evaluate whether a suggested move is better than the one you wanted to play.

You also need to be aware of my background: I am not a super-strong player. I don’t play a lot of human OTB games, and have a USCF rating in the mid 1600’s. My online ratings range from about 1500 (lightning) to 1900 (standard). So I have enough knowledge to be dangerous. Therefore, I invite any comments and criticisms of this article.

Problem areas

Blocked/positions While a computer plays very well in open positions, it is not always so for closed ones. In these positions, the ability to make long term plans is important, and chess engines currently lack this ability. For additional clarification, take a look at the recent Fritz/Van-Wely game from the Dutch Championship. (Note: I’m not trying to pick on Fritz here, it was just a recent, convenient example):

Trapped pieces: I have observed that some engines are willing to trap a piece, especially when there is the possibility of play against an opponent’s weakness. In the PGN file, I have a recent example of a game I played against La Dame Blanche (LDB). While it can be argued that not a super strong engine, I have observed similar behavior in other more modern engines as well. Incidentally, I like playing LDB, since I can occasionally win a game – I would recommend that you take a look at it.

Sacrifices for positional compensation: (this includes attacks) A little while ago, I was going through a book by Leonard Shamkovich called "The Modern Chess Sacrifice". This excellent book discusses positional sacrifices in depth, and is very instructional (sorry, it’s out of print too). However, I noticed that when I followed the games using several of my computer chess engines for analysis, they missed some of the sacrificial moves discussed in the book. A more recent example would be the DB vs. Kasparov game (see PGN file).

Many endgame positions: It turns out that endgame knowledge can be extremely difficult to encode into a chess engine. It is very possible that there may be some blind spots in the particular engine you are using. There are quite a large number of rules (and exceptions to rules) that one can use to evaluate a position. The problem with encoding all these rules into the evaluation of the chess engine is that the engine gets bogged down. Also, don’t forget that you may able to visualize a long sequence of moves quite well, but the engine has to calculate these. And it takes longer and longer for the engine to calculate with each move that it has to look ahead. Here is an example that was recently brought to my attention by Fernando Villagas (see the PGN file).

Tendencies

Predispostion to opening of the position: Given the choice between an open and closed position (with all other things being equal), most engines prefer the open position. Be aware of this when analyzing your games.

Propensity for pawn/piece exchanges: Since computers tend to calculate forcing sequences very well, many also like to exchange pawns/pieces. Clearly, this may not be called for in the position you are looking at, or it may run contrary to your particular playing style. Some engines seem more likely to go for piece exchanges than others, so you will have to make a judgement based on which engine you are using.

Good areas

Thematic pawn moves and any kind of freeing move. If there is a freeing pawn move that can be made in a position, you can bet that your chess engine will probably find it. As I said elsewhere in this article, computers prefer open positions, so there are typically some 'bonuses' assigned when the engine sees that it can open things up.

Tactical or forcing sequences: I won't belabor the point here. Today's engines do not miss a tactical trick. If there is a tactical resource in a position, your engine will find it. Many engines use special "search extensions", meaning they look more deeply at forcing sequences. The end result is that they do not miss much at all.

Play in open positions: While computers are not that good in closed positions (relatively speaking), they really shine in open ones. And this is why they tend to steer the game into these channels.

Short plans or play for static positional advantage: By this I mean the transfer of a knight to an outpost square or build up of rooks on an open file. Most engines are capable of implementing these short plans. Very long term plans are another matter entirely.

Defending tactical play/creation of threats and counterplay: Once things get down and dirty in an attack, you can count on a computer to weasel it's way out. They are also tenacious defenders, and excel at counterattacks.