No comments

MadChess 3.2.1 Released

  By

I have released version 3.2.1 of my chess engine.

This release fixes a bug that caused MadChess to report illegal principal variations. That is, PVs that contain illegal moves. To be clear, the bug never caused MadChess 3.2 to crash. Merely to report fantasy lines.

I didn’t notice the issue back in January because of my desire to transition from programming chess to playing chess- which involves fighting the urge to tinker with one more engine issue. I did wonder why MadChess occasionally reported truncated PVs in the CuteChess GUI. One factor contributing to my (chosen) ignorance is the CuteChess GUI (which I use to run tournaments so I can watch games in progress) silently truncates PVs that contain illegal moves. Whereas the CuteChess CLI (which I don’t use because I cannot watch games in progress) displays a warning. Despite my instinct that a bug infested the PV code, I decided not to go down that rabbit hole until this weekend.

This release does not add any features nor playing strength. Testers may substitute this version for version 3.2 without any need to restart or re-run tournaments.

Thanks to Steve who commented on the MadChess 3.2 Released post that CuteChess CLI complains of illegal PVs.

You may download x64 and x86 versions of MadChess 3.2.1 from the Downloads page. Install the appropriate version for your computer’s CPU. The x64 binary is the strongest version of the engine.

3 Comments

MadChess 3.2 Released

  By

I have released version 3.2 of my chess engine.

The main feature in this release is a sophisticated limit-strength mode, including adjustments to chess knowledge, search speed, chance of erring or blundering, and severity of blunders. MadChess, in limit-strength mode, excludes moves it considers unreasonable (dumb moves even a chess novice wouldn’t play). I have extensively documented the settings and technical implementation of MadChess’ limit-strength mode in The MadChess UCI_LimitStrength Algorithm. I also provide a Limit-Strength FAQ.

MadChess 3.2 is slightly stronger than the prior release. I estimate it has gained 60 rating points, climbing to roughly 2770 Elo at bullet chess.

This likely is the last release of MadChess- at least for a long while. I plan to focus more on playing chess instead of coding chess. And I plan to move on to other (non-chess) programming projects. You may follow along on my general programming blog, ErikTheCoder. Nothing new there yet. But I intend to resume blogging there soon. I’ll keep blogging here also, focused as I said on playing the game, not coding it.

You may download x64 and x86 versions of MadChess 3.2 from the Downloads page. Install the appropriate version for your computer’s CPU. The x64 binary is the strongest version of the engine.

Enjoy!

Follow the discussion of MadChess 3.2 in the TalkChess forum.

No comments

Never Resign!!

  By

The Moral Lesson

I love this video. We’ve all been there, so we can relate. We’ve all screwed up. You don’t have to play chess or even understand the rules of the game to know exactly how Alexandra Botez feels.

Chess is a very difficult, humbling game. But that’s the point. If we only attempted to do what was easy, we never would grow and mature as chess players, as human beings.

I love the raw energy of this video: Alexandra Botez’s obvious frustration. Her cathartic scream…

Fuck!!

How she wards off tilt and steels her nerves. Her determination to play on and make her opponent prove they can win. Her war cry of…

Never resign!!

… when she regains the initiative. The anxiety-inducing ticking of the clock in a one minute bullet game with no increment. (I don’t know how people can play 1 minute bullet. I have enough difficulty managing my time in 3 + 2 blitz .) Her ultimate triumph with only 0.4 seconds left on her clock.

Plus the backing music sounds like an outtake from Metallica’s Death Magnetic album.

The Game

The game begins 1.Nf3 f5 2.d3 Nf6 3.g3 e6 4.Bg2 Be7 5.O-O O-O 6.Nbd2 d6 7.c3 e5 8.e4 fxe4 9.Nxe4 Nxe4 10.dxe4 Qe8 11.Be3 Qh5 12.Qd2 Bh3 13.Ne1 Nd7 14.Bxh3 Qxh3 15.Ng2 Nf6 16.f3 a5 17.Bg5 Rae8 18.Bxf6 Rxf6 19.Qe3 Rh6 20.Kf2 b6 21.Rh1 d5 22.exd5 and arrives at this first critical position.

The game continues 22…Bc5 23.Rae1 Bxe3+ 24.Rxe3 Rf6 25.c4 Qf5 26.Rhe1 Qc2+ 27.R1e2 Qxc4 28.Rd2 Rd8 29.Rxe5 Rfd6 30.Ne3 Qc5 31.Ke2 h6 32.Rc2 Qd4 33.Re4 Qf6 34.Rd2 c6 35.Rf4 Qg6 36.Rg4 Qe8 37.Kf2 cxd5 38.Nf5 R6d7 39.Nxh6+ Kh7 40.Nf5 d4 41.Rh4+ Kg8 42.Rg4 d3 43.Re4 Qf7 44.Rg4 Kh7 45.Nh4 g6 46.Ng2 Kg7 47.Nf4 Qf6 and arrives at a second critical position.

Alexandra Botez captures her opponent’s queen to even the score- actually she’s up one pawn. Capitalizes when her opponent hangs a rook. Then wins two pawns, promotes a pawn to a queen, drives her opponent’s king to a corner, and maneuvers her rook and queen to deliver mate on the back rank a split-second before her time runs out. Amazing!

I gave my chess engine, MadChess (rated 2700 Elo = super Grandmaster), one minute to analyze the game. I also gave one minute to a world-class engine, Komodo Dragon (rated 3500 Elo = much stronger than the best human player). It seems appropriate to give the computers the same amount of time the human players had for the game. You can play through the game below, review suggested improvements by MadChess and Komodo Dragon, and read a few comments and explanations I added.

If you’ve never played chess online, and this video piques your curiosity about what you’re missing, consider signing up at chess.com. Don’t worry, you can play slower-paced games (or games with no clock) than what you see in the video above. You’ll be matched with players of a similar strength as you.

No comments

Testing Strength Reduction Parameters

  By ,

I played an entertaining game against MadChess a few evenings ago. Prior to the game, I adjusted MadChess’ strength reduction parameters because I felt their values caused the engine to play too strongly for a given Elo rating. Perhaps my adjustments made the engine too weak. I’m using “feel”, a very unscientific process. Nevertheless, I enjoyed the game.

I played white. MadChess played black, set to 900 Elo. Time control is blitz, 5m + 5s. The game began 1.e4 c5 2.Nf3 g6 3.d4 d6 4.dxc5 Qa5+ 5.Nc3 Qxc5 6.Nd5 Qc6.

Here I missed a tactic. Can you spot it?

I didn’t see it. The tactic is available on this move and my next three moves because MadChess didn’t “see” the tactic either. See Search Speed for an explanation of how MadChess’ strength reduction algorithm affects the engine’s “sight”.

The game continued 7.Be2 e5 8.O-O h6 9.Be3 Be7 10.c4 Kf8 11.b3 Be6 12.Rc1 Na6 13.Qd2 Qc8 14.Ne1 Bd8 15.Nd3 Qd7 16.f4 exf4 17.N3xf4 Kg7

r2b2nr/pp1q1pk1/n2pb1pp/3N4/2P1PN2/1P2B3/P2QB1PP/2R2RK1 w - - 1 18

Let’s light this fuse. Bd4+!

Material is even, but clearly I’m winning. However, winning “won” games at blitz time control is not guaranteed at my patzer skill level. There’s always a chance I blunder material back to my opponent. Or run out of time. Pushing those concerns aside, I pressed my advantage, didn’t make any egregious errors, and managed my time well.

After I won the game, I gave Komodo Dragon two seconds per position to analyze the game. I then played through the game with Komodo Dragon and MadChess running at full strength, each displaying four best moves (MultiPV = 4). I explored a few variations where I didn’t understand why a particular move didn’t work, consulted the engines, and added refutation lines and comments to the game.

No comments

Banks 96th Amateur Series Division 7

  By

MadChess 3.1 participated in Graham Banks’ 96th amateur tournament in division 7.

MadChess 3.1 won the tournament!

                           1    2    3    4    5    6    7    8    9    0    1    2    
1   MadChess 3.1 64-bit    **** ½0½½ 1½0½ ½½½½ 001½ 1½11 1111 ½011 11½1 ½½11 111½ ½½1½  29.5/44
2   Inanis 1.1.0 64-bit    ½1½½ **** 1½0½ ½1½½ 1001 ½½0½ ½01½ 111½ 01½1 ½1½1 ½111 ½½½1  27.5/44
3   Devel 4.0.2.3          0½1½ 0½1½ **** 0½½½ 010½ ½110 1½01 ½1½1 ½½½½ 1101 1½½½ 11½½  25.5/44
4   Odonata 0.6.2 64-bit   ½½½½ ½0½½ 1½½½ **** ½011 0½½½ 0101 1½½½ ½0½1 1½01 111½ 0011  24.5/44
5   Zevra 2.5 64-bit       110½ 0110 101½ ½100 **** ½0½½ 11½0 ½½1½ ½½1½ 1½½0 ½1½½ 0½½1  24.0/44  522.25
6   Lozza 2.4 64-bit       0½00 ½½1½ ½001 1½½½ ½1½½ **** ½1½½ ½½½½ ½0½1 0½11 ½½10 1½11  24.0/44  500.25
7   Supernova 2.4 64-bit   0000 ½10½ 0½10 1010 00½1 ½0½½ **** ½1½1 ½011 11½½ 10½0 ½½11  21.5/44
8   Blunder 8.5.5 64-bit   ½100 000½ ½0½0 0½½½ ½½0½ ½½½½ ½0½0 **** 111½ ½1½½ 011½ ½111  21.0/44
9   Delocto 200419 64-bit  00½0 10½0 ½½½½ ½1½0 ½½0½ ½1½0 ½100 000½ **** 1½1½ 0½11 ½½1½  19.5/44
10  KnightX 3.5 64-bit     ½½00 ½0½0 0010 0½10 0½½1 1½00 00½½ ½0½½ 0½0½ **** 1½11 ½½11  18.0/44
11  Myrddin 0.89 64-bit    000½ ½000 0½½½ 000½ ½0½½ ½½01 01½1 100½ 1½00 0½00 **** 1½11  16.0/44
12  EggNog 4.0 64-bit      ½½0½ ½½½0 00½½ 1100 1½½0 0½00 ½½00 ½000 ½½0½ ½½00 0½00 ****  13.0/44

Games