ELECTRONIC T-NOTES


CHESSBASE USA'S WEEKLY ON-LINE NEWSLETTER


FOR THE WEEK OF NOVEMBER 21, 1999


THE WONDERS OF BLUNDERCHECK

by Steve Lopez

The purpose of "Blundercheck" analysis in the playing programs we offer is primarily to check analysis given in variations added to a gamescore in .cbh format. You can annotate a game and then have Fritz (or any of the other playing programs/engines) check the variations for errors. However, there's a hidden feature of blundercheck that advanced users will find interesting.

In normal analysis provided by Fritz, it will show you places where a player went wrong with a given move, presenting an improvement and an occasional text comment. Long-time users of the program (going back to Fritz2 and Fritz3) are used to seeing numerical analysis, in which you not only see that a move was not only better but receive a numerical evaluation of the position at the end of the variation line. By using blundercheck, you can get the old-style numerical analysis with an extra feature: Fritz will also give you a numerical analysis of the move that was actually played in the game so that you can compare it with the move the program suggests.

Click the Database button in Fritz5.32 (or any of the other playing programs we offer, of course), highlight the game you wish to have analyzed, and click the "Analyse" button. You'll be given the choice of "Full analysis" or "Blundercheck". Select "Blundercheck" and you'll see the following dialogue box:

You can choose to have the moves of only one player analyzed. I generally choose to have both players' moves examined in order to see any mistakes my opponent made that I may have missed.

You can choose either "Time" or "Depth"; these are mutually exclusive (you can't select both). I'll typically choose "Time" and set it for 60 to 90 seconds. Play around with these settings to see what works best on your machine. If you choose "Time", this is an average length of time per move. It's not etched in granite. If you choose 90 seconds, for example, Fritz will get to 90 seconds, finish the ply it's currently searching, and then move on. Note that if you have a very fast machine, Fritz may get into double-digit ply numbers by the time 90 seconds is reached. If the curent ply results in a very large search tree at that point, it may take a considerable amount of time for the program to get to the end of that ply. This is why I said you need to play around with the feature and figure out the time setting that works best for you. For example, if I set my (slow) computer to anything higher than 120 seconds, Fritz will typically take 3-8 minutes per move (beacuse it needs to finish the ply it's reached at 120 seconds) and this can tie up my computer for a considerable period, especially on a longer game.

If you select "Depth", try to be reasonable about what you choose. Even on the fastest available PC hardware, it's not likely that the program will go past 15 plies (and this is with a fairly long search -- 20 to 30 minutes per move). If you have a Pentium II 11 or 13 are good values to use in this field. Always use an odd number -- this will help to avoid tactical oversights by making the program always consider the opponent's reply. Using an even number causes the program to sometimes be blind to possible responses.

"Threshold" is a number given in 100th's of a pawn. In the example diagram above, I've set Fritz for 30, which is 0.30 of a pawn. Any improvement it finds which is less than 0.30 of a pawn better than what was actually played will not be displayed in its analysis, while any improvement which is 30/100th's of a pawn or better will be shown. This allows you to set the program for either tactical or positional analysis. If you set a higher number you will see only tactical improvements. For example, setting it for 100 will provide analysis in which only improvements that win a pawn or more will be shown. As an extreme example, you could set it for 300; in such a case only piece-winning improvements will be displayed.

Conversely, if you set "Threshold" for lower numbers, you'll be presented with positional improvements. A value of 10 will show you very Steinitzian improvements in pawn structure, etc. I'm not one for hair-splitting, so I use 30 or 35 for the value -- that way I'll see meaningful positional improvements without getting into the nitty-gritty of tiny incremental improvements.

"Output" allows you to select how the moves are displayed. Long-time users will recall that Fritz2 used to present the variations it found as text rather than as replayable moves. Back in those days I used to have to make a printout of each analyzed game and enter the variations by hand. Selecting "Annotate as variations" will give you the analysis as replayable moves instead of just text. I don't know why anyone would want to use "Annotate as text" but the option is there if you want it (hey, one can never have too many choices!)

"Storage" allows you to select whether the game will be "appended" to the database as a new game at the end or whether the current game in the database will be "replaced" (overwritten) by the Fritz-annotated version. This is a pretty significant choice. If you're using Fritz to analyze a previously-annotated game, you'll probably want to use "Append" so as to not screw up or lose the prior analysis. If the game is unannotated, you can just use "Replace". Important note: neither of these options will work on a game located on a CD! It's not physically possible to write changes to information on a CD, so you'll need to copy the database ot the game to your hard drive first before having Fritz annotate it!

"Write full variations" means that Fritz will show you its full analysis and the numerical evaluations. Leave this box checked. "Erase old annotations" is at your option; obviously, you don't want to select this if you're having Fritz check your added variations for errors! "Training" allows Fritz to add timed training questions to the game to point out and test you on tactical mistakes it finds.

The final two boxes allow you to select whether Fritz will analyze only the main line (the moves actually played in the game), just the variations you've added to the game (a true blundercheck, as described in the manual), or both (if you select both check boxes).

Once you've set the parameters, let it rip by clicking "OK". I usually let this run overnight on a game since an in-depth analysis can take hours. Short (5-10 second per moves) analyses are quick but not usually very helpful, as Fritz doesn't look very deeply into the possibilities.

When you come back the next morning, you'll see that the database (game list) screen is once again displayed on your monitor. This is how you'll know that Fritz is finished with the analysis. Double-click on the game to see what Fritz found for improvements to what was actually played.

Here's an example (from one of my games) of what blundercheck analysis looks like. My opponent (in this case a computer program set to a handicap level) has played a weak move: 14...Ncxe5. This has caused its position to deteriorate quickly in Fritz5.32's view. Fritz says that in this position White is 1.25 pawns ahead. Black could have improved on its game by playing the variation Fritz suggests, after which it would have been only 0.09 pawns behind -- a huge difference (assuming best play for both sides, of course).

You'll notice that the last reading shows "0.09/10". This means that White is 9/100ths of a pawn ahead at the end of the variation shown and that Fritz5.32 did a 10-ply search to arrive at this conclusion.

Instead of getting just a variation with a text comment (as we would have using "Full analysis"), "Blundercheck" shows us not just the improvement but also how much of an improvement it would have been -- in this case 1.16 pawns' worth.

Many players (particularly beginners) won't care for this style of numerical analysis, but more advanced players (particularly the more analytical among us) can benefit greatly from the precise numerical evaluations of the actual and potential positions within Fritz' analyzed games.

Until next week, have fun!

You can e-mail me with your comments, suggestions, and analysis for Electronic T-Notes. If you love gambits, stop by my Chess Kamikaze Home Page and the Yahoo Chess Kamikazes Club.