Random Posts

Play Live Blitz


Friday, June 8, 2018

Chess Engine Review

     I recently received an email from a reader wanting to know which engine they should use for analysis. So, here is a post about engines that I hope will answer the reader's question. 
     Engines make our lives easier...just load a game and let the engine work out the lines. For those that want to improve, this approach can't be recommended. For best results you have to analyze on your own and only then compare your analysis to the engine's. Even then, it won't tell you why one move is better than another unless there is an immediate tactical refutation. 
     Engines aren't concerned with anything more than what move yields the highest possible evaluation and they don't care about traps and dangerous counterplay. So, as far as humans are concerned, in a lost position the best move may not be the one with the highest engine score; it may be that there is a move that doesn't score as well, but offers the opponent more chances to go wrong. Another case where you may not want the best move is in a situation where a secondary move gives the easiest win. The recent post of the Cochrane – Mohishunder game is a good example. 
     Sometimes it can be dangerous to trust the evaluation of engines. A position evaluated at 0.00 doesn't mean it's a draw. It may mean that the chances are even, but in order to keep the balance you might have to play 10 perfect moves. Try that sometime! In other cases, it may show one side has a huge advantage, but in the long run it may not be able to win. And, in some positions the complications may be easy for an engine to see through but not a human. 
     As readers who play through the games in this blog know, I sometimes refer to Shootouts. I run Shootouts when I am not sure I believe the engine's evaluation or don't seen how the win can be accomplished. I discussed Shootouts in THIS post. As I pointed out in that post, it's not the individual moves I am interested in so much as the general direction the games took as that can be very helpful in finding a plan. 
     As far as I know, the Fritz GUI is the only program that has the Shootout feature. The newest version of the Fritz program costs about $70. I've seen some places where you can supposedly download some earlier versions for free, but it's probably illegal and dangerous, so I wouldn't do it. Fortunately, my old Fritz 12, which I found several years ago at Office Max for $20 (it was selling online for three times that), is still working and it runs Komodo and Stockfish just fine. 
     At the moment the top three rated engines on CCLRs 40/40 list are: 
1-Stockfish 9 64-bit 4CPU (3444) 
2-Komodo 12 64-bit 4CPU (3412) 
3-Houdini 6 64-bit 4CPU (3409) 

     There are two key elements in engine analysis: search and evaluation. Search is the way that the engine prunes the tree of analysis. Pruning is necessary because because with each ply the number of possible moves grows exponentially and so some moves have to be trimmed in order to obtain greater search depth. Evaluation is the set of criteria used by the engine to evaluate a position. 
     The main difference between engines is in their evaluation function. Stockfish seems to be best in the endgame and in spotting very deep tactics. Komodo is best at evaluating middlegame positions accurately once the tactics are resolved. Houdini is the best at blitz and at seeing tactics quickly. This explains why the very best correspondence players use different engines and often look at moves that aren't in the engine's top choices. 
     One nice feature of Chess Assistant is that you can refer to its list of all legal moves and tell it which ones you want it to analyze.   I don't know of any other program that has this feature. In order to analyze a specific move with other programs you have to play it then let the engine do its analysis. In some positions that can be very time consuming. 
     Stockfish is very aggressive in the way that it prunes its analysis, so it can search very deep but as it goes forward, it searches fewer plies. This is advantageous in the endgame and in some sharp tactical positions, but it can be a disadvantage in other positions. 
    Because Komodo's consultant, Larry Kaufman, was a Grandmaster its evaluation is the most positionally accurate. This is important when there is a material imbalance or the position is closed. It's also the best engine for playing the opening when out of book early. The engine is slightly slower than either Stockfish or Houdini and it seems to need longer analysis time than do the other two. 
     Houdini best for tactics, Stockfish for endgames and whenever great depth is required. Komodo has the best sense for relative piece values, I think. Both Houdini and Stockfish overvalue the Queen.  
     Houdini is a tactical whiz that tends to do best on the various tactical test sets that some engine experts have put together, and it spots tactics fairly quickly. One important point has to be made concerning Houdini's evaluation numbers. They are designed to predict the outcome. 
     A half Pawn advantage gives a 50 percent chance of winning. A one Pawn advantage gives an 80 percent chance of winning. And, a three Pawn advantage will win 99 percent of the time. There is one caveat though...those odds are against an equal opponent at blitz time control. Your results may be different!
     The question is, which engine is best? It depends. Any engine will point out our tactical mistakes, even second tier engines like Crafty and Fritz. But, if you are doing serious analytical work or the position is very complicated you need all three because each one of these engines has their strengths and weaknesses.
     Houdini and Komodo will cost you, but for most of us the free versions will do just fine. The free Houdini 1.5 can be downloaded from the Houdini site HERE. Komodo 9 can be downloaded free on the Komodo site HERE. Stockfish 9 can be downloaded HERE
     In the following game the position after black's 14th move is quite interesting. Caruana gave his 15th move two "!!" and in his notes wrote that no engine would find it. That was four years ago and out of curiosity I analyzed the position with Stockfish 9, Komodo 10 and Houdini 1.5, giving each engine five minutes.
     The results confirm that, as mentioned, an evaluation of 0.00 doesn't always mean a draw and sometimes that from the human perspective, there may be a more promising move than the engine's recommendation. 

Top two moves: 
Stockfish: 15.Na4 (0.16) and 15.Nd6 (0.00) 
Komodo 15.Na4 (0.01) and 15.Nd6 (0.00) 
Houdini 15.Nd6 (0.07) and  15.Na4 (0.06) 

After 15.g4: 
Stockfish (-0.07), Komodo 10 (-0.08), Houdini (-0.13) 

    Clearly, there's not much difference between the engines' top two choices and Caruana's move, but his move gives white more practical chances. Kingscrusher (real name Tryfon Gavriel, owner of www.chessworld.net and FIDE Candidate Master) analyzes the game. 

No comments:

Post a Comment