Random Posts

Thursday, July 14, 2016

Shootouts in Unclear Positions

     Exactly what is an "unclear" position? The answer itself is unclear. What's unclear to me is likely to be very clear to a GM. What's unclear to a garden variety GM may be very clear to a super-GM. 
     In books, especially older ones, when you come across the term "unclear" in describing a position it usually means a) the annotator didn't feel like analyzing the position or b) he had no idea what the correct evaluation is, or c) the position really is unclear because both players have so many ideas that it's impossible to choose the best one. 
     With the help of engines there are a number of analysis tools you can use and in Fritz these are the Deep Position Analysis option and the Shootout option. I rarely use the DPA option though it has its advantages. The most important is that it will analyze moves that at first evaluation do not look too good, but may turn out to be best. Mostly I don't use it because getting good results takes time and most of my correspondence play is in "rapid" events where the time limit is 10 days plus 1 day per move and no vacation time is allowed. Most games are finished in 3-4 months. I just don't want to take the time to use DPA. 
     I love the Shootout feature though. It's especially helpful when an engine can't seem to come up with a clear evaluation. You start with a position of interest and then get one or more engines to play the position with set times or with increasing search depths. A couple of things to note: 1) in some positions where the evaluation is nearly equal, after running Shootouts the move under investigation actually does poorly and 2) at lower plies one side does well, but at the higher plies things may change. 
     While analyzing one of my games on LSS recently the engines (Stockfish and Komodo) evaluated the position as almost dead equal and recommended three moves that were separated by only a couple hundredths of a point. 
     The position and actual scores are not so important, but my best three moves were:

16...a4 (This move made no sense to me. Was, in the absence of any tactics, the engine just making aimless moves? They do that sometimes.) 
16...Bf5 (Developing a piece and attacking the white Pawn on d3, but it also allowed him to play Bxb7 attacking the R on a8.) 
16...Rd8 (This move attacked the white B that was shielding his d-Pawn and so seemed quite reasonable. And even though it was evaluated a wee bit lower than the other two I was tempted to play it.) 

     After a lot of tinkering with the position I wasn't able to come to any clear conclusion, so Fritz in the Shootout mode came to the rescue. I could have used the Deep Position Analysis to generate a tree of possibilities, but chose the Shootout option, letting Stockfish and Komodo play out the game. As usual, the ply settings were 13-19 plies which results in five games being played by each engine. 
     In the Shootout mode, when the games are finished the results will give you a pretty good indication of the likely outcome; in my game it will probably be a draw. After running a Shootout I play through the games quickly and look at the way the engine played. This gives me a hint of available ideas and to how the position should be played. By the way, this information is also useful for opening analysis, especially if you are doing research for OTB play. 
     The Shootout results were as follows (white's results are shown): 

16...a4 
Stockfish +0 -0 =5
Komodo +0 -0 =5 
16...Bf5 
Stockfish +1 -0 =4 
Komodo +1 -0 =4 
16...Rd8 
Stockfish +1 -0 =4 
Komodo +0 -0 =5 

    In the two 16...Bf5 games which white won it was because he got a better K position and his B was more active than black's in the ending. 
     In the case of 16...Rd8 white got a mobile Q-side majority and black was fighting to hold the draw. 
     So, the seemingly pointless (to me) 16...a4 was the best move after all, but I was curious as to why this was so. Playing through the games, I discovered the reason was that it allowed for the possibility of activating the a8R by the maneuver ...Ra5-b5 attacking the b-Pawn which tied up one of white's B as well as one of his Rs. This maneuver was not possible without ...a4 because a R-lift with the moves ..Ra3 and ...Rb3 was not playable. 
     It was not so much the individual moves that I found instructive, but the ideas that were apparent from the general direction the games took.  
     That's why I like doing Shootouts when things are not as clear to me as the engine evaluations make it seem and it's why you will occasionally see Shootout results in my game annotations.

No comments:

Post a Comment