In the previous blog post we mentioned the question, “What are the best analysis settings for the Fritz family of chess playing programs?” (said programs being Fritz, Hiarcs, Junior, Rybka, and Shredder). Today we’ll consider the answer to that question.
And, spilling the beans early, I’ll tell you straight away: there are no “best” analysis settings.
The settings and parameters you use will, by and large, depend on your needs and personal preferences, as well as on your level of development as a chess player. There are certainly bad settings which should be avoided (and we’ll discuss those, too), but there’s really no “optimum” setting for parameters such as “Threshold”, nor is one type of analysis (Full analysis or blundercheck) objectively “better” than the other. It all depends on the individual user’s requirements.
As I mentioned previously, “Full analysis” mode is likely the better choice for less experienced chessplayers, due to the relative brevity of the alternative variations it provides, the short verbal commentary, and the fact that variations are evaluated symbolically rather than with numerical values.
Here’s an example of a game analyzed using “Full analysis” mode:
and the same game analyzed with “Blundercheck”:
As a chessplayer who’s also been involved with computer chess for nearly twenty years, I’m very comfortable with Blundercheck (and, in fact, I prefer it) because it gives me much more “hard” information than does “full analysis”. But it’s also easy to see how some players could be overwhelmed by the torrent of numerical data provided by Blundercheck mode. The core purpose in both forms of analysis remains the same: to show where each of the players went wrong and what they should have played instead. “Full analysis” mode provides the information in a much simpler form.
Let’s narrow it down to a particular part of the game to examine the differences in closer detail:
In this view from the full analysis mode, we see the variation to Black’s 22nd move starts with the symbol for “Better is”, but the variation for White’s 21st move doesn’t. This is important, because full analysis will occasionally show variations which aren’t improvements in an effort to answer the “But what if ___ had been played?” questions that novices often ask. The variation to White’s 21st move is provided to answer that “what if…?” question. The variation at Black’s 22nd move, on the other hand, shows a better plan of action; even though Black is already winning by that point in the game, the move 22…Qg4 would possibly have led to a faster, more efficient victory.
Compare the “Blundercheck” output:
Note that we don’t get an alternative move for White at move 21; “Blundercheck” mode will only show improvements, not “what if…?” moves which are actually worse. As for Black’s 22nd move, we get additional information which “Full analysis” didn’t give us: a longer variation, plus a numerical evaluation for both the variation and the move which was actually played in the game. This is absolutely the entire point of using “Blundercheck”. You’ll recall that Black is completely winning at this point in the game. A player reading this analysis might be thinking: If Black’s winning anyway, then what’s the point of a “better” line at move 22? “Blundercheck” answers this question with complete precision, as we’ll see…
Note the numbers after White’s 22nd move, as well as Black’s. After 22.Nxg5, Black is winning by 3.73 pawns – about three and three-quarters pawns (you’ll recall that positive numbers mean an advantage to White, while negative numbers indicate an advantage for Black). Now look at the number after 22…Qh5 (highlighted in the illustration); with that move, Black’s advantage drops to 1.60 pawns – less than one and two-thirds pawns. It’s readily apparent that 22…Qh5 was not the move that Black should have played. Now look at the Black variation which begins with 22…Qg4. Had Black played that move, and assuming best play for both players (indicated by the variation provided), after 30…h3 Black would have had an advantage of 3.73 pawns – exactly the evaluation provided for the position after White’s 22.Nxg5. What the numbers are telling us, described verbally rather than numerically, is that Black would have retained his full advantage by playing 22…Qg4; with 22…Qh5 he gave up some of his advantage (although he’s still winning, but another similar slip-up might give White a chance to crawl back into the game).
Which analysis form you use is ultimately your choice. If you’re interested only in where you went wrong (but not by how much) and in seeing suggested improvements, you might prefer “Full analysis”. If you’re interested in far more precise information about game errors, and don’t want the alternate “what if…?” variations which aren’t actual improvements, then you’d likely prefer “blundercheck”.
As far as “Time” settings (regardless of which analysis mode you use) are concerned, here again you’ll find that there are no “best” settings. There are so many variables at play: availability of RAM for hash tables, processor speed, number of processors (for “Deep” versions of ChessBase playing programs) and (most important) how much time you have available for a game’s analysis. But you need to remember the cardinal rule: the higher the “Time” setting, the better the analysis you’ll receive in return. There is, however, a law of diminishing returns which you also need to bear in mind. On modest (by current standards) hardware, a chess engine will achieve a search depth of 12-14 plies nearly instantaneously. As more plies are layered on, the analysis process begins to slow down until at some later point (dependent on the hardware variables mentioned above) a single extra ply can take many minutes (or even hours) to complete. The whole trick for you is to find that “sweet spot” for your hardware configuration in which you’re getting a decent analysis depth without tying up your computer for days on end.
One thing I will state absolutely, however (since I said we’d talk about bad settings), is that 5 seconds a move for a “Time” setting is far too short and, in fact, I wouldn’t recommend any setting less than 30 seconds as a rock-bottom low end “Time” setting. I know that there are users who want Fritz, etc., to analyze their complete corpus of hundreds of games played over years of tournament and club competition in under a minute total (for all the games, not per game), primarily because I have occasion to talk to these people, or correspond with them from time to time. This stuff ain’t Star Trek, gang; despite the speed of present-day home computer hardware, it does take a little time for chess software to analyze a chess game. The analysis shown in the first two illustrations above took about an hour per game to generate (achieving a typical search depth of seventeen to twenty plies) on a fairly modest single-processor home PC at sixty seconds a move. Not instantaneous, but danged impressive nevertheless.
There is a trick, by the way, for setting up your computer to analyze multiple games, which we will discuss (along with the subject of “Threshold” settings) in our next post this coming Tuesday. Until then…
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.