Lately I’ve been playing more frequently against the computer program Shredder, with a purpose. After my last tournament I had a feeling that I was not getting enough practice in winning superior positions. In fact, playing against Shredder set at its maximum strength, I was hardly ever even getting a superior position. So I came to the conclusion that I needed to play some training games against the computer with its rating “dumbed down” a bit.
In retrospect, it’s silly that I didn’t do this before. There were two main reasons. One was pride. I wanted to vanquish the computer in a fair fight, or not at all. The second was that I expected the weakening algorithm to work in a very artificial way. I figured that the computer would play the best move 9 times out of 10, and then on the 10th move (or whatever) it would play some crummy move because the program said that it was time to play something stupid.
However, in practice it hasn’t felt that way at all. It has actually felt more like playing a human opponent. If I fight really hard and don’t back down, I’ll win a few. I get a chance to play combinations and sacrifices, which almost never happened against full-strength Shredder. If full-strength Shredder allows a sacrifice, it’s almost certainly unsound. After a while that gets discouraging and I stop looking for such moves.
I’d like to show you a game I played today against Shredder at a time control of game in 10 minutes. It was the highest rating level I have beaten it at so far (I set its rating to 2316). The game was of particular interest to me because it’s one of my favorite opening variations and one where I have a prepared improvement.
Shredder — Dana
1. e4 e5 2. Nf3 Nc6 3. Bb5 Nd4 4. Nxd4 ed 5. O-O g6 6. d3 Bg7 7. Nd2 Ne7 8. f4 c6 9. Bc4 d5 10. Bb3! …
FEN: r1bqk2r/pp2npbp/2p3p1/3p4/3pPP2/1B1P4/PPPN2PP/R1BQ1RK1 b kq – 0 10
Position after 10. Bb3. Black to move.
I know I’ve written about this variation before, but let me do it again. 5. … g6 initiates the Blackburne (Sub-)Variation of the Bird Variation. I’ve gotten to the position after move 9 several times in tournaments, and in all but one game my opponents have played 10. ed? They most likely think that by doing this they are giving me a wretched pawn structure, but what they’ve really done is suck all the dynamic possibilities out of White’s position.
The correct move 10. Bb3! was played by Jose Raul Capablanca against Joseph Henry Blackburne in St. Petersburg, 1914. That was about the last time this position was seen in top-level chess, so convincingly did Capablanca win. But I have an improvement over Blackburne’s 10. … O-O.
10. … h5!
Black does not intend to castle! At least not until much later. Instead I plan to play … Bg4 and exchange my bishop for the only minor piece that can really do damage to me on the kingside. I also have designs on attacking on the kingside myself, with the rook remaining on the h-file.
I have not yet gotten to try this out in a tournament. Unfortunately, Capablanca is no longer playing chess (at least on Earth).
11. Nf3 Bg4 12. ed Nxd5 13. Re1+ Kf8!
This always surprises people, but it’s a characteristic maneuver in the Blackburne Variation.
14. h3 Bxf3 15. Qxf3 Bf6?! 16. f5! …
The most enterprising move. Arguably I should have played 15. … Qd7 or 15. … f5 on the last move to discourage it.
16. … g5 17. Bd2 Qd6 18. Re4 …
Okay, we’re due for another diagram, and besides, this is a key position.
FEN: r4k1r/pp3p2/2pq1b2/3n1Ppp/3pR3/1B1P1Q1P/PPPB2P1/R5K1 b – – 0 18
Position after 18. Re4. Black to move.
Here I was starting to get the feeling that White was taking over the game. There is one good way for Black to keep that from happening. This being a ten-minute game, I didn’t have enough time to be sure that my pawn sacrifice was sound, but I knew that if I didn’t play it my position would start going downhill.
18. … g4!
I don’t know whether Shredder “allowed” this because I had set its rating to 2316, or whether White legitimately had nothing better than to allow this move. For what it’s worth, I put the game on Rybka (full strength) and it liked Shredder’s previous move 18. Re4, too. So maybe Black is just plain equal. Which is great news!
19. hg hg 20. Rxg4 …
A human might have declined the pawn sac, but computers generally expect you to prove them. By the way, White could also have taken with the queen. I did a lot of analysis on 20. Qxg4 with Rybka, and I think it might have been a better choice for the computer from a practical point of view. The best (and only) equalizing moves for me would not have been checks: 20. … Ne3 and 20. … Rh4. Those would have been harder moves to find. (Note that Black does not want to play 20. … Qh2+ because there are mating threats for White if the knight should move or be taken.)
By contrast, in the 20. Rxg4 variation I can reel off four straight checks before I have to think again.
20. … Qh2+ 21. Kf2 Bh4+ 22. Ke2 Re1+ 23. Re4 Rxe4+ 24. de? …
This was White’s losing move. I imagine that full-strength Shredder would not have made this mistake, but 2316 Shredder, playing with a 10-minute time control, optimistically thought the position is +1.34 pawns in its favor! According to Rybka, White can hold after 24. Qxe4 Nf6 25. Bf4! Bg3! 26. Bxg3 Qxg3 27. Qf3! with only a token 0.16-pawn advantage for Black.
FEN: 5k1r/pp3p2/2p5/3n1P2/3pP2b/1B3Q2/PPPBK1Pq/R7 b – – 0 24
Position after 24. de. Black to move.
Don’t touch that knight! Remember what I said: The only way to beat the computer is to never compromise and never give in. Actually, the winning move is not all that hard to find:
24. … Rg8!
White simply has no way to defend g2.
25. Bh6+ Ke8 26. ed Rxg2+ 27. Kd3 Rg3 28. Re1+ …
I was a little surprised by this; I was expecting 28. Qxg3, when it looks as if maybe White can get to a playable endgame (R + B vs. Q). But the computer is very good at working out all the tactics, and it sees that in all variations Black can either mate or win more material.
28. … Kd8 29. Re3 …
Definitely not a human move. The computer is programmed to minimize the damage, even if it means giving up a whole rook. If 29. Rf1 Qg2! White is also losing a ton of material.
29. … de 30. Qe2 … (diagram)
FEN: 3k4/pp3p2/2p4B/3P1P2/7b/1B1Kp1r1/PPP1Q2q/8 b – – 0 30
Position after 30. Qe2. Black to move.
Okay, so here I have to make a confession. Shredder is a very gentlemanly opponent, and it will give you a chance to take back a move if it sees that it is a serious blunder. Here I originally played 30. … Qxe2+, which is okay but it leads to an endgame where Black still has a lot of work to do before he can win. I was very surprised when Shredder gave me the message, “I think that move was not good,” and asked if I wanted to take it back. That gave me a clue: I shouldn’t just settle for trading queens. Given that clue, it was pretty easy to see what I should play here:
30. … Rg2!
Of course! White can’t take on e3 because the pin happens all over again. For what it’s worth, Rybka comes up with a cleverer way to trade queens: 30. … Qf2! If White trades, Black queens a pawn. Otherwise, 31. … Qxf5+ is murderous. However, the text move is straightforward and effective.
Here Shredder played 31. Bf4, and now I was glad to trade queens because I also get a bishop. After 31. … Rxe2 32. Bxh2 Rxh2 the computer could have resigned, but it played on for 32 uninteresting moves before resigning.
Aside from a bit of an asterisk on move 30, I don’t think that there is anything about this game I would change if I were playing at a full tournament time control!
{ 6 comments… read them below or add one }
Interesting post. I am also interested in playing against an engine for training as well due to fact I don’t get chance to play OTB. Can I ask what/which platform? PC? tablet? iPad?
Do you leave it to random openings or do you specify openings? if latter, can I ask how/what you do?
I’m not the greatest authority, but I can tell you my experience. I switched over from the PC/Windows world to the Mac world in 2010. I was disappointed to see that there were not many options for top-level chess programs on the Mac. Shredder 12 was the best choice I could find. Meanwhile, I kept a toe in the PC world with a laptop, and I downloaded Rybka 3 in about 2011. I use Shredder exclusively for training, and Rybka for analysis. (I prefer not to play on the laptop because its screen is so small.)
Shredder plays a great variety of openings and seldom plays the same variation twice. This is great if you want to broaden your opening knowledge, not so good if you want to practice a specific variation. However, you can if you want enter the position that you want to play using Analysis mode, then switch over to Play Against Computer mode.
Whether you’re playing them at top strength or at a lower strength, computers don’t play quite like humans. I think that playing a computer is great for sharpening your tactical awareness — better even than playing a human. For practicing your strategic chess and for learning to deal with human psychology (your opponent’s and your own), training against a computer is not so good.
But if there aren’t any human players where you are, what other options do you have? There’s online chess, but I am not a fan of that because you don’t really know whom or what you are playing, and because there are annoying issues of sportsmanship (or lack thereof) that turned me off.
I use my chess engine to practice opening play. I play it at max strength up to 12 moves. By the 12th move it’s usually obvious I’ve found a hole in my opening knowledge.
What do you mean “…computers don’t play quite like humans”? Please elaborate.
Interesting game. Thanks for sharing. I am not a strong chess player but I am a decent programmer. I’ve written a chess engine that includes support for UCI_ELO & UCI_LimitStrength, an interface for weakening a chess engine, supported by the Shredder application.
I see in your comment above that you’ve switched to Macs. My program runs only on PCs. And it maxes out at about 2250 ELO- but that’s against other chess engines. Against humans it’s probably rated higher, as its results on ICC suggest. But I thought I’d mentioned it here in case you want to check it out.
I’m unsure how Shredder weakens its playing strength, but I can speak to how I accomplished it in my chess engine. I had the same concern as you. I didn’t want my engine to play strongly for 9 moves, then play a ridiculous 10th move. I’ve written a detailed explanation of the weakening algorithm on my website. Hopefully understandable to a layperson with a science background.
http://www.madchess.net/page/The-MadChess-UCI_LimitStrength-Algorithm
ChessBase sell a couple of CDs called Fritz Technique Trainers by Henrik Schloessner. The CDs supply graded sets of game positions that Schloessner considers winnable for humans playing against Fritz. The idea is to practise winning won positions. I can see there might be some value in this: learning how to sit on opponents and limit their counterplay. I suppose it can’t be that difficult to collect a few suitable positions and give it a whirl. But I haven’t got around to it yet…
A bit late: many thanks for your Bird Defence series!