ELECTRONIC T-NOTES


CHESSBASE USA'S WEEKLY ON-LINE NEWSLETTER


FOR THE WEEK OF AUGUST 8, 1999


USING YOUR COMPUTER AS A RESEARCH ASSISTANT (PART TWO)

by Steve Lopez

Last week in ETN, we took a look at the position that occurs after the following moves:

1.e4 e5 2.Nf3 Nc6 3.Bb5 f5 4.Nc3 fxe4 5.Nxe4 d5 6.Nxe5 dxe4 7.Nxc6 Qg5 8.Nd4+ c6 9.Bf1+

I decided to use Mega Database 99 as my reference database and created an opening report using ChessBase 7. The preliminary findings indicated that the line's not bad for White, even though it's not a standard "book" variation.

The opening report suggested three likely candidate moves for Black in this position as well as replies for White, and offered some statistical information based on the thirteen games in which the move 9.Bf1 appears in the Mega Database.

However, there are a couple of extra features that the "Opening report" command provides which we didn't examine last week. The first of these is the window called "Repertoire for printing". This provides a nice ECO-style table that you can print out in order to have a hardcopy of the games the opening report turned up.

The initial window looks like this:

What ChessBase has done in this case was merge the thirteen games found by the opening report into a single game. Notice that the "Table" tab is automatically selected at the top of the notation window. This gives an ECO-style table of the variations. The table can be scrolled left to right by using the scroll button at the bottom of the notation window. Alternately, one could click the "Diagram size" button at the bottom of the game window, select "None", and get a wider version of this table:

A third way to display the moves is to click "Full" at the top of the notation window to get the standard-style gamescore with alternate moves appearing as variations:

This repertoire table (or game) can be printed out. To get a table-style printout, select "Repertoire" from the Printing menu. To get a standard game-style printout, select "Print game" from the Printing menu.

Either way, once you've printed it, you're no longer chained to your computer. You can take the printout with you wherever you go and do some analysis on a pocket chess set in your spare moments.

The other window that the opening report provides is a statistical tree generated from the games found when you created the report:

The tree initially appears with the position you searched for visible on the chessboard. But you can easily step backward through the tree to the initial position for a game of chess if you so desire. There's no way to save this tree to disk, so you must use it immediately or else recreate it later; if you close it, it's gone.

Every caveat I gave last week concerning the use of statistics in the opening report applies to this tree as well. Thirteen games is not a large enough statistical sampling to put a lot of stock in the figures given in this tree. The primary use of the tree is to make it easy to step through the variations and branches of the game to see what's been played by masters in these positions.

Last week, we saw that the opening report alerted me to the possibility of 9...Qe4 as my opponent's reply, followed by 9...Bg4 and 9...Bc5 (in order of preference). In this tree, we also see two more moves: 9...Nf6 and 9...Bd6. Each was played in just a single game. In the game I played against a computer program (the game which prompted all of this analysis), my silicon opponent played 9...Bd6, so this move immediately catches my eye. Stepping ahead one move in the tree, I see that White's reply was 10.d3, which is what I played in my game. My computer opponent played 10...Qe5, but the only move in the tree is 10...Qa5+, which was easily countered by 11.Bd2, which was followed up by 11...Qe5. So in effect, we get the same position as in my game with White having had an extra move. So perhaps my computer opponent had more on the ball than I'd initially suspected.

Stepping back through the tree to the position after 9.Bf1, I decide to explore the possibilities after 9...Qe5 (the move deemed by the opening report to be the most likely to appear). The tree shows three choices for White: 10.c3, 10.Nb3, and 10.Ne2. I don't much care for 10.c3; advancing this pawn will make d2-d3 more difficult later and I'd like to be able to use that move to pressure the e4-pawn. 10.Nb3 looks like a turkey as well; as discussed last week, it takes the Knight out of the action. 10.Ne2 looks to be the way to go. However, the statistical tree shows that the move was only played once and that the game was a win for Black. What gives?

First of all, the move doesn't immediately lose. It's not an instant blunder, so the reason for Black's victory must have come later in the game. This is why you can't let statistics run your game for you. If you played strictly by statistics, you'd have to play 10.Nb3 here (with 100% success for White) and I don't like the Knight's offside placement at all. Your mileage may vary, of course, but I'd rather keep the Knight closer to the center.

After 10.Ne2, the tree shows 10...Bc5 (which I think is a bit questionable), followed by 11.d4 (with a nice fork of the Black Bishop and Queen, albeit one that's easily countered). Black strikes back with 11...exd3 (now we see another reason for 10.Ne2 -- it shields the White king from the Black Queen), and White replies with the obvious 12.Qxd3 (since 12.cxd3 gives White an isolated d-pawn).

Note that I'm evaluating these positions myself, instead of letting a computer program do it for me. By analyzing the positions "by hand", I'm actually learning a great deal about them which can only help me in future games. If I had a playing program do the analysis, I'd get finished a lot faster but I wouldn't understand a tenth as much as what I do now ("Oh, OK, Fritz says White's OK here -- on to the next position", and twenty positions later I still wouldn't understand a dang thing).

But how do I double-check my thinking in these positions? This is where a playing engine comes in. Rather than using an engine to just barf out reams of analysis, I use an engine as a tool to double-check my thinking about each of these positions after I've already done the skull sweat.

Let's consider an example. Stepping back through the tree, I come to the position where Black has played 9...Qe5 in response to 9.Bf1. After looking at the tree and considering my options, I decided that 10.Ne2 was the best move (regardless of what the statistics said). Notice that there is a set of buttons at the bottom of the tree window. One of them is the "Start/Stop analysis module" button that so many players have come to rely on. Clicking this button provides a list of engines that can be used for positional analysis. I can use these engines to double-check my critical thinking in these positions.

Before we go on to look at some specific positions, there are a couple of things to keep in mind. Users frequently ask me how long they should let an engine analyze before considering the results. This will tend to vary depending on your computer hardware. Be aware that every extra ply you add to a search increases exponentially the number of positions that must be searched compared to the previous ply. Consequently the search depth you'll be able to reach depends a lot upon your processor speed and the amount of RAM you've allotted for hash tables. As a practical example, I can do a nine-ply search with Fritz5.32 in under a minute on a Pentium II 266 MHz computer. Another minute or two will get me into the eleventh ply, but then things start slowing down. If I'm willing to go out for a walk or go watch some TV while the program searches, I can get a search through fourteen complete plies and partway into the fifteenth ply within twenty to thirty minutes. For most positions, though, I'll let Fritz do a complete eleven ply search (meaning that the display will show that it's currently working on the twelfth ply), which usually takes about five to eight minutes on my machine. If I'm feeling especially lazy or if the position is not too tough, I'll do a nine-ply search instead.

Another factor to keep in mind is that you should let the computer analyze an odd number of plies. If you have it look at an even number (eight, for example) it will look at four moves for each side. But when it finishes the eighth ply, it will then be the original moving side's turn to make a move again. It's better to do an odd number -- that way you'll get an extra ply's analysis for the side whose turn it is to move in the initial position (the selective search function of chess engines minimizes the risk of missing a devastating reply by the opponent in these cases).

Going back to the position after 9...Qe5, I rev up Fritz5.32 and let it rip. I've set the analysis display to show the four best moves that Fritz is considering. As we've seen, of the thirty-two possible moves in the psition, only three make a lick of sense. However, the results are not what I thought they'd be. The top move is evaluated as giving White a distinct advantage (+/- in Informant symbology), while the second and third choices show White with a slight edge (+/=). And there's a good bit of numerical variance between the three lines of play that Fritz is displaying:

10.Nb3 (0.72)
10.c3 (0.56)
10.Ne2 (0.38)

There's a pretty hefty difference in the evaluations between the top move and my third-place move (+0.34 -- a third of a pawn).

So it looks like Fritz and I are at odds in this position (you'll recall that I selected 10.Ne2). Looking ahead at Fritz' preferred line of play, I see that after 10.Nb3, Fritz is considering 10...Nf6 11.Qe2 Bd6 12.d4 to be the best play for both sides. This makes sense; Nb3 keeps e2 open for the Queen, plus the Knight influences the d4-square. Qe2 puts pressure on the isolated e4- pawn and the move d4 kicks the Black Queen (plus the pawn is protected by the b3-Knight). Additionally, Black can't play ...exd3 e.p. because he would then lose his Queen to Qxe5.

This is pretty smart tactical play and they're moves I'd not even considered. Looking at the second-best choice, Fritz sees 10.c3 (supporting the Knight), 10...Bc5 (hitting the Knight a second time), 11.Nb3 (dropping back to support a pawn advance to d4), 11...Bd6 (avoiding the coming fork), 12.d4 exd3 e.p.+ 13.Be3 as the likely continuation.

What about my preference (10.Ne2)? It's not bad, but doesn't look as good to Fritz as the other possibilities. 10...Nf6 11.d3 exd3 e.p. 12.Qxd3 Bf5 13.Qb3 and now the White Queen is off to the side (where I didn't want the Knight to go).

So now I'm crying for a second opinion, since different chess engines will evaluate positions differently (just as strong chessplayers do -- Karpov and Kasparov are two top-level players with radically different styles of play; chess engines are no different). I stop Fritz5.32 and let Junior5 have a whack at the position.

Junior's style is more cautious than Fritz'; while Fritz is a furious attacker, Junior tends to play a bit more tentatively, getting its own house in order before flying into an attack. This is reflected in Junior's analysis of the position, in which all three variations give White a slight edge (+/=), but their ranking is different from Fritz':

10.c3 (0.68)
10.Nb3 (0.54)
10.Ne2 (0.28)

There's a 0.40 spread between the top move and the third move. Junior prefers 10.c3, followed by some fairly quiet play: 10...Bc5 11.Nb3 Bd6 12.a4 Be6 13.d4 Qd5. Junior is suggesting that White should try to gradually get some extra space while Black should move his Bishops off of the back rank.

My preferred move still comes in third: 10.Ne2 Nf6 and Junior ends its analysis there (evidently there are two or more eleventh moves for White that lead to equal evaluations).

Shall we consult Hiarcs7.32 for a third opinion? It's easily the most positionally-oriented of the engines available for use within ChessBase so it will be interesting to see what it comes up with.

10.c3 (0.55)
10.Nb3 (0.29)
10.Ne2 (-0.09)

Ouch! My move is nearly two-thirds of a pawn worse than Hiarcs' choice for the best move. It prefers 10.c3 Nf6 11.Qb3 Bc5 12.Bc4 Bxd4 with an early exchange of Bishops in the center.

My move is a lot less dynamic according to Hiarcs: 10.Ne2 Bc5 11.Nc3 Nf6 12.Qe2 0-0 13.f4 Qd4 and Black gets a pretty good grip in the center.

Meanwhile Crafty16.13 also concurs with the other engines that my choice is third best. It chooses 10.c3 as the best move (rating it numerically at 0.43). My move is given an evaluation of 0.32. Crafty doesn't see as large a gap between the three moves, but it's still clear that my choice is not the best.

I guess I'll have to rethink this position. I'll examine all four programs' analyses and decide which one makes the most sense to me (as well as which one best suits my style of play). And this is a key point: I'm not taking the engines' analyses at face value, but instead taking the time to examing the suggested moves in an attempt to understand why a particular engine thinks a move or line is better than what I came up with. Chess engines are strong players but they're not perfect, so you need to determine for yourself whether or not the moves they come up with are better than your owns ideas and (most importantly) to determine the reasons why.

All of this is an example of how we can use a chess engine (or set of engines) to double-check our evaluation of a given chess position. We can also use engines to double-check statistical results. Note that in the initial examination of the tree the move 10.Ne2 had a 0% success rate for White. However, three of the chess programs that looked at the position said that 10.Ne2 still gave White a slight advantage, while one of them showed the position as equal. It's pretty safe to assume that White made some poor moves later in the game and that's what cost him the point. Of course, I still want to double-check it by playing over the game in question.

In my analysis earlier in this article, I raised some questions about other positions. I can use the various chess engines to double-check my analysis there, too (but I won't burden you with the analysis).

Now let's return to the position that started all the furor: 9.Bf1. Is this an OK position for White? The results from master practice seem to indicate that this is the case. Again, we can use the chess engines to check this. Let's see what our silicon masters have to say:

Fritz5.32 thinks that this position is all right, giving White a slight edge after 9...Qg6 (it evidently doesn't like having the Queen on the same diagonal as the c1-Bishop, because of the threat of 10.d3 [with a discovered attack on the Queen plus an attack on the e4-pawn]) 10.c3 (guarding the Knight and creating a pocket on c2 for the White Queen) 10...Bg4 11.Qc2 0-0-0 12.f3 exf3 13.Qxg6 hxg6 14.gxf3 and Black hasn't proved that his gambit was a success.

Junior5 also thinks that White has a slight edge here: 9...Nf6 10.Qe2 Bd7 11.Nb3 Bd6 (notice that Black is concentrating on development in this line. This makes sense, since the usual idea in gambit play is to sacrifice a pawn to obtain fast development) 12.Qc4.

Hiarcs7.32, too, thinks that there's a slight edge for White here: 9...Qe5 10.c3 Bc5 11.Nb3 Bb6 12.Bc4 Nh6 (headed for g4 with the double attack on the f2-pawn upcoming).

And Crafty16.13 also tips the scales slightly toward White after the interesting variation 9...Qd5 10.c3 Nf6 11.f3 Kd8 (a very curious and puzzling move. Perhaps Crafty is thinking ahead toward 12.fxe4 Nxe4 and wants to avoid a pin on the Knight after 13.Qe2 -- it's hard to determine what it's thinking here) 12.d3 exf3 13.Qxf3 Bg4 14.Qe3.

Going over these variations and analyzing the reasons for these moves, plus determining the reasons why these variations are better than my own ideas has helped me to understand the initial position much more thoroughly than if I'd simply had an engine or two chew on the position with no input from me. By directing the analysis and taking careful note of the results, I now know that 9.Bf1 is playable for White and that my idea for White's tenth move isn't necessarily bad -- it's just not as good as 10.Nb3 (and, best of all, I know why this is so).

There are some other ways to use playing programs to analyze positions. If you have Fritz/Nimzo/Junior/Hiarcs you can use the "correspondence analysis" feature to generate a whole tree of analysis from a particular position. This is time-consuming, however, often requiring hours to create an analysis tree using just a single engine but providing a very thorough picture of the possibilities that can arise from the position you're studying.

Another idea is to play games against a chess engine starting from the study position. For example, I could play a few games against Fritz5.32 starting from the position after 9.Bf1, allowing Fritz to play the Black pieces. Then I can change engines and play a few more games against another engine that has a different playing style (such as Hiarcs, Nimzo, or Crafty). This is a good way to get some practical experience in that position against a variety of "opponents". Of course, you'll want to have your chessplaying program analyze the game afterwards to show you where you could have improved your play (and the "Comparative analysis" feature in our playing programs allows you to have more than one engine analyze a game so that you can compare their suggestions).

Don't be afraid to analyze master games using chess engines. I've frequently mentioned that a great way to understand an opening variation is to play through as many games as possible that contain it. But sometimes you'll get stuck. For example, I might play through the 10.Ne2 game from my Mega Database and find myself stumped as to where White went wrong. In that case, I can have Fritz5.32 or any of the other engines analyze the game overnight and show me how White lost the game (probably through a series of weak moves rather than a single major blunder).

And finally there's always the idea of using books in conjunction with your computer software. Even though my position after 9.Bf1 wasn't a "book" move, I still found it worthwhile to check into various explanatory texts to discover the ideas behind the Schliemann Defense. Fine's Ideas Behind the Chess Openings, Horowitz' Chess Openings: Theory and Practice, Schiller's Standard Chess Openings and Unorthodox Chess Openings, and Kallai's Basic Chess Openings and More Basic Chess Openings are just a few of the books that provide this kind of background material. (I've written a large number of chess book reviews that can be found on-line at my Chess Kamikaze Home Page; you'll find a link to it at the end of this article).

In these two articles on using your computer as a research assistant I've given you a lot of ideas to try out. I'm sure that you can discover many ideas of your own in addition to the ones I've presented. Just remember the key point: do your own thinking! You'll sometimes be wrong in your evaluations (as I was in my initial examination of 10.Ne2), but you're sure to learn a lot about the positions you're studying by analyzing them yourself rather than just falling back on software tools to do the analysis for you. Every position you analyze and understand increases your chess knowledge. The more you know, the more links and connections you create in your brain and the easier future analysis tasks will become. You may be analyzing a position that you'll never see again, but some portion of what you've learned will stay with you and will be applicable to similar positions you'll see later. Doing the analysis yourself first is a much better learning experience than simply having a program do it for you.

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 the Chess Kamikaze Home Page and the Yahoo Chess Kamikazes Club.