Your trusted source for computer chess information!
Differences
This shows you the differences between two versions of the page.
| Both sides previous revisionPrevious revisionNext revision | Previous revision | ||
| computer_chess:engines:myrddin:chesswarxiv [2009/05/29 17:53] – jvmerlino | computer_chess:engines:myrddin:chesswarxiv [2009/05/31 18:45] (current) – jvmerlino | ||
|---|---|---|---|
| Line 33: | Line 33: | ||
| Analyzing this game gave me several things to work on, primarily checking the algorithms for giving appropriate bonuses and penalties to castling, as well as investigating the King Safety code (which is still pretty poor in Alpha 2). The moves 7.gxf3 and 9.Rg1 are pretty bad, when 7.Qxf3 and 9.dxe6ep are obviously much better (Myrddin Alpha 2 now plays 7.Qxf3, but prefers 9.f4?, which is at least a slight improvement). 11.Rb1 is also quite difficult to justify. The four straight bishop moves from 13 through 16 are pathetic and accomplish very little.\\ | Analyzing this game gave me several things to work on, primarily checking the algorithms for giving appropriate bonuses and penalties to castling, as well as investigating the King Safety code (which is still pretty poor in Alpha 2). The moves 7.gxf3 and 9.Rg1 are pretty bad, when 7.Qxf3 and 9.dxe6ep are obviously much better (Myrddin Alpha 2 now plays 7.Qxf3, but prefers 9.f4?, which is at least a slight improvement). 11.Rb1 is also quite difficult to justify. The four straight bishop moves from 13 through 16 are pathetic and accomplish very little.\\ | ||
| - | After 18.Bh3 (Myrddin Alpha 2 will still play this move), I am regretting allowing Myrddin to enter the tournament. It was provisionally rated by Olivier at 1699 (rather high, in my opinion -- possibly by as many as 300 points for the initially entered version), but so far this is 1200 play at best. I made many improvements to the engine over the one month of the tournament, adding things that most other engines rated in Myrddin' | + | < |
| + | # FEN | ||
| + | r3k3/ | ||
| + | </ | ||
| + | r3k3/ | ||
| + | |||
| + | With 18.Bh3 (Myrddin Alpha 2 will still play this move), I am regretting allowing Myrddin to enter the tournament. It was provisionally rated by Olivier at 1699 (rather high, in my opinion -- possibly by as many as 300 points for the initially entered version), but so far this is 1200 play at best. I made many improvements to the engine over the one month of the tournament, adding things that most other engines rated in Myrddin' | ||
| + | < | ||
| + | # FEN | ||
| + | r3k3/ | ||
| + | </ | ||
| + | r3k3/ | ||
| Finally, Usurpator makes a slight blunder (24...Nxh4), | Finally, Usurpator makes a slight blunder (24...Nxh4), | ||
| Line 107: | Line 119: | ||
| Pooky finished in 45th place with 7 points.\\ | Pooky finished in 45th place with 7 points.\\ | ||
| - | This was Myrddin' | + | This was Myrddin' |
| So much for ProDeo' | So much for ProDeo' | ||
| 5...Qa5+ is a patzer move if there ever was one. 9...Bg7 and 10...0-0 were a surprise, as I had never seen Myrddin follow through on a fianchetto before.\\ | 5...Qa5+ is a patzer move if there ever was one. 9...Bg7 and 10...0-0 were a surprise, as I had never seen Myrddin follow through on a fianchetto before.\\ | ||
| + | < | ||
| + | # FEN | ||
| + | r4rk1/ | ||
| + | </ | ||
| + | r4rk1/ | ||
| + | |||
| 18...Ne5 is dreadful, and that kind of move is still a problem with Myrddin Alpha 2. It simply can't see that, after all of the potential captures are played out, the knight on b4 is lost. 18...a5 is the only move. From here, the game should be lost.\\ | 18...Ne5 is dreadful, and that kind of move is still a problem with Myrddin Alpha 2. It simply can't see that, after all of the potential captures are played out, the knight on b4 is lost. 18...a5 is the only move. From here, the game should be lost.\\ | ||
| - | But Pooky gives the full point back wth 26.Rfe1. I can only assume that it had some serious bonus for the loss of two pawns in front of the enemy King. After 34...Rxd5, material is even, and the game looks very drawish, although Myrddin has the passed pawn.\\ | + | < |
| + | # FEN | ||
| + | r5k1/ | ||
| + | </ | ||
| + | r5k1/ | ||
| + | |||
| + | But Pooky gives the full point back with 26.Rfe1. I can only assume that it had some serious bonus for the loss of two pawns in front of the enemy King, to make up for the loss of the Bishop. After 34...Rxd5, material is even, and the game looks very drawish, although Myrddin has the passed pawn.\\ | ||
| + | < | ||
| + | # FEN | ||
| + | 8/ | ||
| + | </ | ||
| + | 8/ | ||
| With 62...Rf3+, somehow Myrddin finds the right plan and ends up with the only pawn on the board. When the pawn queened on move 82, it felt pretty good. The fact that Myrddin played optimum moves through the end of the game, without EGTBs, was also satisfying. \\ | With 62...Rf3+, somehow Myrddin finds the right plan and ends up with the only pawn on the board. When the pawn queened on move 82, it felt pretty good. The fact that Myrddin played optimum moves through the end of the game, without EGTBs, was also satisfying. \\ | ||
| Myrddin now has two points in two games, and I'm very encouraged.\\ | Myrddin now has two points in two games, and I'm very encouraged.\\ | ||
| Line 157: | Line 187: | ||
| Bad luck with the matchup, as Myrddin ends up on the top board against the eventual winner of the tournament. Still, playing top board is another nice moment in my first tournament.\\ | Bad luck with the matchup, as Myrddin ends up on the top board against the eventual winner of the tournament. Still, playing top board is another nice moment in my first tournament.\\ | ||
| Myrddin plays book moves up through and including 7.Kxd1, but I am not amused. Yes, Myrddin is up a pawn, but failing to castle is not a good thing for Myrddin.\\ | Myrddin plays book moves up through and including 7.Kxd1, but I am not amused. Yes, Myrddin is up a pawn, but failing to castle is not a good thing for Myrddin.\\ | ||
| + | < | ||
| + | # FEN | ||
| + | 2k4r/ | ||
| + | </ | ||
| + | 2k4r/ | ||
| + | |||
| 19.Nf4 is the losing move, although after the exchange of two pieces for a rook and pawn, Myrddin gave an eval of practically a draw (0.01 for White). A typical engine gives the position at least +1.5 in White' | 19.Nf4 is the losing move, although after the exchange of two pieces for a rook and pawn, Myrddin gave an eval of practically a draw (0.01 for White). A typical engine gives the position at least +1.5 in White' | ||
| Myrddin doesn' | Myrddin doesn' | ||
| Line 220: | Line 256: | ||
| Finally a game in which Myrddin gains quite a bit of time at the opening, as Piranha is out of book five moves before Myrddin.\\ | Finally a game in which Myrddin gains quite a bit of time at the opening, as Piranha is out of book five moves before Myrddin.\\ | ||
| After the first two dozen moves, Myrddin just hasn't accomplished much at all, and Piranha has all of the space. 24...h5 is silly.\\ | After the first two dozen moves, Myrddin just hasn't accomplished much at all, and Piranha has all of the space. 24...h5 is silly.\\ | ||
| - | 26...Kf7 is just not the way to protect the e-pawn. 26...Nc5 | + | < |
| + | # FEN | ||
| + | 2r2rk1/ | ||
| + | </ | ||
| + | 2r2rk1/ | ||
| + | |||
| + | 26...Kf7 is just not the way to protect the e-pawn. 26...Nc5 | ||
| While working on Myrddin for the next round, I realized that I had broken my 3-fold repetition check. Nor was I able to find out why before the next round started. I realized that I unfortunately had to send in an earlier version (the one before round 2) in order to " | While working on Myrddin for the next round, I realized that I had broken my 3-fold repetition check. Nor was I able to find out why before the next round started. I realized that I unfortunately had to send in an earlier version (the one before round 2) in order to " | ||
| Line 253: | Line 295: | ||
| Alice finished in 111th place with 5.5 points.\\ | Alice finished in 111th place with 5.5 points.\\ | ||
| - | A very painful result, as Myrddin played acceptably well up until 24.Qxa6, allowing the draw. Once I was finally able to fix the repetition checking code, Myrddin Alpha 2 avoids this move instantly, and instead chooses the correct Rc1. Sigh.... \\ | + | < |
| + | # FEN | ||
| + | 3r3N/ | ||
| + | </ | ||
| + | 3r3N/ | ||
| + | |||
| + | A very painful result, as Myrddin played acceptably well up until here (up an exchange with a good attack), where it played | ||
| At least Myrddin still has scored 50% through the tournament. My goal was to have it place in the top 70 out of 212, and before this game Myrddin is ranked slightly below that mark. Since it was initially seeded 37th, this doesn' | At least Myrddin still has scored 50% through the tournament. My goal was to have it place in the top 70 out of 212, and before this game Myrddin is ranked slightly below that mark. Since it was initially seeded 37th, this doesn' | ||
| Finally I was able to fix the repetition checking code, plus add pondering. It turns out the problem with the 3-fold check was that I was being inconsistent as to when I was generating the board signature for moves. During the search, I was correctly generating a signature before making the move. But when I received moves from the opponent, I was generating the signature AFTER making the move. This is a very good way for the whole process to fail.\\ | Finally I was able to fix the repetition checking code, plus add pondering. It turns out the problem with the 3-fold check was that I was being inconsistent as to when I was generating the board signature for moves. During the search, I was correctly generating a signature before making the move. But when I received moves from the opponent, I was generating the signature AFTER making the move. This is a very good way for the whole process to fail.\\ | ||
| Line 289: | Line 337: | ||
| Simon finished in 85th place with 6 points.\\ | Simon finished in 85th place with 6 points.\\ | ||
| - | Myrddin is winning out of book, although the evals don't show it. Myrddin plays several sub-optimal moves, but never gives up its advantage, although 14.Kb1 is definitely odd (Myrddin Alpha 2 would now play the acceptable Bc4). Simon' | + | Myrddin is winning out of book, although the evals don't show it. Myrddin plays several sub-optimal moves, but never gives up its advantage, although 14.Kb1 is definitely odd (Myrddin Alpha 2 would now play the acceptable Bc4). |
| + | |||
| + | < | ||
| + | # FEN | ||
| + | r4k1r/ | ||
| + | </ | ||
| + | r4k1r/ | ||
| + | |||
| + | The game is already lost here, but Simon' | ||
| After this game I implemented LMR (Late Move Reductions) and got a significant speed increase. \\ | After this game I implemented LMR (Late Move Reductions) and got a significant speed increase. \\ | ||
| Line 334: | Line 390: | ||
| I was very concerned about this game; just had a bad feeling. Aldebaran ended up with the same score as Myrddin, but had a much easier time of it. So, due to tiebreakers, | I was very concerned about this game; just had a bad feeling. Aldebaran ended up with the same score as Myrddin, but had a much easier time of it. So, due to tiebreakers, | ||
| Myrddin had a small advantage after the opening, but for some reason decided to trade pawns rather than develop.\\ | Myrddin had a small advantage after the opening, but for some reason decided to trade pawns rather than develop.\\ | ||
| + | < | ||
| + | # FEN | ||
| + | 2r1k2r/ | ||
| + | </ | ||
| + | 2r1k2r/ | ||
| + | |||
| 17...Bb4 is a very poor move, and Myrddin Alpha 2 will still make the same move today. Thankfully, Aldebaran failed to see that the knight sac 18.Nxd5 exd5 19.e6! wins.\\ | 17...Bb4 is a very poor move, and Myrddin Alpha 2 will still make the same move today. Thankfully, Aldebaran failed to see that the knight sac 18.Nxd5 exd5 19.e6! wins.\\ | ||
| + | < | ||
| + | # FEN | ||
| + | 4k2r/ | ||
| + | </ | ||
| + | 4k2r/ | ||
| + | |||
| Myrddin maintains a slight advantage up through 27...e3, although oddly enough both programs think that White is winning (Aldebaran gives +0.94 and Myrddin gives +0.55). I suspect most top engines would give Black about +0.30 advantage. After 27...e3 (which Myrddin Alpha 2 still plays), Aldebaran now has the slight advantage, which will get bigger as Aldebaran is up a pawn and several pieces are going to come off the board.\\ | Myrddin maintains a slight advantage up through 27...e3, although oddly enough both programs think that White is winning (Aldebaran gives +0.94 and Myrddin gives +0.55). I suspect most top engines would give Black about +0.30 advantage. After 27...e3 (which Myrddin Alpha 2 still plays), Aldebaran now has the slight advantage, which will get bigger as Aldebaran is up a pawn and several pieces are going to come off the board.\\ | ||
| Then Myrddin gets its second big break of the tournament, as Aldebaran fails to make the time control while thinking about its 40th move.\\ | Then Myrddin gets its second big break of the tournament, as Aldebaran fails to make the time control while thinking about its 40th move.\\ | ||
| Line 403: | Line 471: | ||
| Zotron finished in 76th place with 6 points.\\ | Zotron finished in 76th place with 6 points.\\ | ||
| - | Myrddin, for the second time in the tournament, makes a losing trade of two pieces for rook and pawn with 11...Bxc2. I would assume that this can be solved by slightly increasing the value of bishops and knights, but Alpha 2 still has an even 3 points for these pieces. But it's really 15...d4 that is a killer. Myrddin Alpha 2 would now play the somewhat reasonable 15...Qh4+ in this position. Apparently getting the passed pawn into enemy territory was judged as valuable, but definitely not for the eventual loss of two pawns. Things get progressively uglier from here. By move 50, Zotron is up a piece and a pawn, and about 15 moves later, Myrddin has a lone king. Game over, right?\\ | + | < |
| - | Well, apparently not, as Zotron must have had some code that said "if I make a move and my opponent has no legal responses, that's good!" However, Zotron' | + | # FEN |
| + | r2qk2r/ | ||
| + | </ | ||
| + | r2qk2r/ | ||
| + | |||
| + | Myrddin, for the second time in the tournament, makes a losing trade of two pieces for rook and pawn with 11...Bxc2. I would assume that this can be solved by slightly increasing the value of bishops and knights, but Alpha 2 still has an even 3 points for these pieces. | ||
| + | |||
| + | < | ||
| + | # FEN | ||
| + | r2qk2r/ | ||
| + | </ | ||
| + | r2qk2r/ | ||
| + | |||
| + | But it's really 15...d4 that is a killer. Myrddin Alpha 2 would now play the somewhat reasonable 15...Qh4+ in this position. Apparently getting the passed pawn into enemy territory was judged as valuable, but definitely not for the eventual loss of two pawns. Things get progressively uglier from here. By move 50, Zotron is up a piece and a pawn, and about 15 moves later, Myrddin has a lone king. Game over, right?\\ | ||
| + | |||
| + | < | ||
| + | # FEN | ||
| + | 8/ | ||
| + | </ | ||
| + | 8/ | ||
| + | |||
| + | Well, apparently not, as Zotron must have had some code that said "if I make a move and my opponent has no legal responses, that's good!" However, Zotron' | ||
| === Round 9 v Protej 0.5.5 === | === Round 9 v Protej 0.5.5 === | ||
| Line 495: | Line 584: | ||
| Myrddin' | Myrddin' | ||
| Unfortunately SnailChess does not output any PVs, so it's hard to know when it came out of book.\\ | Unfortunately SnailChess does not output any PVs, so it's hard to know when it came out of book.\\ | ||
| + | < | ||
| + | # FEN | ||
| + | 8/ | ||
| + | </ | ||
| + | 8/ | ||
| + | |||
| SnailChess wins a pawn in exchange for Myrddin' | SnailChess wins a pawn in exchange for Myrddin' | ||
