A couple of blog posts back we looked at two separate game analysis forms in the Fritz “family” of chess playing programs (Fritz, Rybka, Hiarcs, Junior, and Shredder), and we learned some important points about the analysis time settings as well. Today we’re going to examine the “Threshold” setting, and learn how to get Fritz, etc., to analyze more than one game in a session.
Let’s look at that little trick for multiple game analysis first. Normally (as you know) you just single-click on a game in the game list to highlight it, then click the button for the analysis form you want to use (“Full analysis” or “blundercheck”). But you can get your playing program to analyze multiple games by highlighting more than one. Here’s what your game list might look like after single-clicking one game:
But let’s say you want to analyze four games which all appear together in the database. How do you highlight them? Just single-click on the first game, hold down the “Shift” key on your keyboard, then hit the “Down cursor” (down arrow) key (while still holding down “Shift”) until all four games are highlighted:
…and then you’ll just click the button for the annotation form you want to use, set the parameters you want the chess engine to follow, and let ‘er rip.
What if the games you want to have analyzed aren’t conveniently located sequentially in the database? You can highlight these, too – just hold down the CTRL key while you single-click on each game you want your chess engine to analyze:
This allows you to “skip” games in the database’s game list. After they’re highlighted, click the button for the analysis form you want, etc.
Now we’ll look at the “Threshold” setting which, you’ll recall, allows you to set a numerical value, expressed in increments of 1/100th of a pawn:
To give an example, let’s say that you set the threshold to a value of “100”. Every time Fritz (or any other chess engine) finds a variation which results in a position which evaluated as better than a full pawn over what was actually played, it will display the variation and evaluation at that point in the game notation:
In this position, even though Black is ahead by nearly 2/3rds of a pawn after 19…g5 (the highlighted move), Fritz has found an improvement. Look at the variation beginning with 19…Ref8. Had that variation (which assumes best play for both sides) been played out to completion, Black would have been ahead by 1.64 pawns (remember – negative numbers mean that Black has an advantage).
One move later, White blunders horribly:
With the move actually played, White will end up more than a minor piece behind positionally (a later variation [not shown] reveals that with best play Black will be up materially a minor piece for a pawn and with a better position). But had White instead played 20.Rb3, and assuming best play for both sides, White would still be behind, but only by about six-tenths of a pawn instead of more than three pawns; White’s position would be more than two pawns better.
With a threshold setting of 100, Fritz will display a variation if it results in a position that’s better than one pawn (or its equivalent) for the moving side. However, if an alternative move/variation results in a improvement of, say, eight-tenths of a pawn (which is still pretty significant, by the way) it won’t show the better move/variation because 80/100ths of a pawn is lower than the “Threshold” value of “100” (e.g. 100/100ths of a pawn).
That’s why the “Threshold” value you set is important.
Now I know what some readers are already thinking: “Then why not set the threshold value to something really low, like ‘0’ or ‘5’?” Well, there’s a good reason why you don’t want to do that – you’ll get a ton of analysis which is pretty superfluous. If you play, say, the second-best move in a position, a move which is only 8/100ths of a pawn “worse” than the best move, you’re still playing pretty dang well. Receiving an engine-generated variation which is that close to the best move amounts to “splitting hairs”. It’s not bad enough a move for any human player to be able to see any real difference, plus that variation’s going to clutter the game notation and possibly cause you to miss/overlook significant problems with what you actually played.
This points out a fact which appears to some to be kind of a paradox, bu makes perfect sense when you know how the “Threshold” value works: the higher the “Threshold” value, the less analysis you receive in return; conversely, the lower the “Threshold” value, the more analysis you get back (but, as I argued in the last paragraph, the less valuable that analysis potentially can be).
Now it’s obviously your software and you set the analysis parameters in any way you choose. But over the years I’ve been involved in computer chess, I’ve found that the “threshold” setting should be tailored to your individual skill level. It’s not likely that a beginner is going to understand a variation which results in the loss of anything less than a pawn (such variations are where we enter into the realm of positional analysis a la Steinitz), which is why I recommend that beginners or low-rated players use “100” as a threshold setting, to see variations which show how the concrete loss of material could have been avoided.
As your chess knowledge improves and you’re able to recognize positional weaknesses in pawn structure and piece placement, you can reduce the “threshold” value accordingly. My preferred setting is “30”, as the loss of 30/100ths (3/10ths) of a pawn is roughly equivalent to the loss of a tempo. As always, your mileage may vary, but I think that “25” (maybe “15” as an absolute minimum) would probably be the lowest recommended setting for all but the strongest players among us. Anything lower and the improvement in an individual variation will often be undetectable by the eyes of mere mortals, plus you’ll experience the “clutter” of gobs of variations in the game notation, making it tougher to pick out the more noticeable, and thus more easily correctable, mistakes.
Have fun! – Steve Lopez
Chessplayers who have purchased their ChessBase brand chess computer software from USCFSales can receive free technical support and advice on their purchases straight from me; just shoot me an e-mail (firstname.lastname@example.org), but please remember to include the USCFSales order number from your ChessBase software purchase. – Steve
Copyright 2011, Steven A. Lopez. All rights reserved.