< Earlier Kibitzing · PAGE 7 OF 7 ·
|Dec-07-13|| ||SChesshevsky: < Jim Bartle: Give any 2600 a knight advantage against a computer with a 32-piece tablebase (in other words, a perfect chessplayer), and he or she could draw or win the great majority of the time.>|
A knight is significant odds to beat but I'm wondering, given that the computers are programed to play the best moves to gain an advantage I guess, could they be adapted to play the best moves to always draw?
Exchanging pieces is great with an advantage but exchanging all the pieces being a Knight down ends up drawing. I remember seeing grandmaster games where a player was a N or B down but was able to remove the opponents last pawn to ensure a draw.
Personally I think its bad enough computers can beat pretty much everybody now but if they can also be assured of a draw even a piece down, that would really ....
|Dec-07-13|| ||Daisuki: <SChesshevsky>, yes, if you exchange <everything> then king and knight vs. king is a draw. Perhaps my use of the word piece was unclear. By default piece means just queen, rook, bishop, or knight. So kings (which can't be traded, of course) and pawns are not [unqualified] "pieces". They are chess pieces, of course, but they're a bit different from the other pieces. Anyway, I mentioned in the past that Carlsen does need to avoid trading off all his pawns. This is generally not a big problem, though.|
As to playing for a draw, I guess the idea would be to avoid tactics that are the best counterplay for engines (who are the "gods" of tactics, but tactics do tend to involve or at least risk material exchanges or sacrifices) and try to go for a super-solid fortress position. I don't think this would work work against good players since this isn't a strategy that seems to be used at the elite level, so they obviously can't make it work against each other (if they could then anyone good enough could have a rating near the top). As I've said, Carlsen just needs to keep two (often just one) pawn exchanges possible and he can position his pieces all he likes before breaking through (while the opponent, being willingly backwards, basically cedes Carlsen's first six ranks to Carlsen), and after that the opponent can't hide anymore, being confined to little of the board as it is.
I think an engine could definitely pull this off at least often enough against me (although I doubt it would be worth it in terms of expected score, as I would basically never lose anymore, since I'd have to blunder heavily in spite of there being no serious tactics to avoid), but I just don't play precise enough, active enough moves to really burn that extra knight's advantage in strongly (this shouldn't be the case for Carlsen, who will cede far less and take far more than I can, given his greater evaluation and especially calculation abilities). Apparently it's good enough to make the engine realize that it can't hold the square gains it made (or just can't make those squares worth anything because it's a piece down). Basically, it must get "excited" about a prospect that would, in a fair game, usually be something it could roll into further pressure and/or material gains, only to later see that, without a material gain, it in fact can't really make progress, because of the unfair extra knight.
|Dec-07-13|| ||Daisuki: My main ideas (which are recommended more for crappier players like myself) so far from the few games I've played against Houdini 1.5a are (in no particular order): |
1. Mirror the opening moves Houdini makes until doing so is potentially bad tactically (or at least until it moves its queen "because" it lacks the queenside knight that it would "normally" move to control the targeted squares). This won't last beyond a few moves anyway, but it does save you time that you can use later to avoid blundering. The moves should be good enough, and it does force Houdini to burn more of its own time earlier. Very early on is a time of less risk for the human, anyway, so this seems fine.
2. Avoid proactively trading pawns; defend pawns attacked by pawns heavily, most importantly with one or even two other pawns (so you can recapture with a pawn if necessary, not a piece), rather than initiating pawn trading. This helps a lot regarding keeping the position closed and thus lowering both the maneuverability of Houdini's pieces and the number of threats Houdini can make per piece, as more of the board is blocked by pawns. This helps you reposition your pieces as well, since you won't initially place them as well as Houdini does.
3. Try to have no more than two pawn islands. This limits the number of open files, which is great because it makes it much easier to trade off your heavy pieces (queen and rooks). It's also important in general because you want Houdini to have as few areas of the board in which it can reasonably try to make progress as possible. The more pawns you have to defend in order to defend the whole pawn island, the more problems you tend to have.
4. Make even trades (i.e. definitely not stuff like three minor pieces of yours for Houdini's queen, nor two minor pieces for rook and pawn) so long as it doesn't hurt your pawn structure and control over important squares too much. "Even" includes trading your bishop for Houdini's knight so long as this doesn't give you an overall color weakness that looks potentially annoying. Trading queens is the best, as queens are dangerous and hard to contain, so perhaps it's worth a <small> concession if you can manage this.
5. Try overdefending squares that are harder to defend once threatened (such as doubled pawns you might allow to open a file on top of getting a trade) so you can move other defenders (such as pawns) elsewhere in the future.
6. Try to always have squares "earmarked" for pieces to retreat (i.e. move back into what is clearly your territory; Houdini's territory is a tactical hell, and the more contested territory can be a pain as well) to so you can reposition them as needed over time.
7. King safety is really important; Houdini doesn't care about sacrificing material to get a draw via repetition, and you in theory don't want this, much less the worse possibility of losing. Avoid having to use your king to defend much (especially more central squares) until you're in the endgame. Watch out for potential forks and pins even if you have those squares locked up by pawns for now. Move your king away and/or overdefend those squares to fix this potential future tactical hazard.
8. Just be patient and conservative overall. If you're not a good player, as I am not, just focus on defending and keeping things solid, while having options to reposition your pieces safely. Eventually more trades will occur and things will get easier. Don't worry about risking Houdini pushing a pawn to its fifth rank; it often will prefer to keep the capture possible as this offers more options, and locking the pawn structure just makes it harder for it to invade with its pieces, so it doesn't tend to take this option (of course letting a pawn advance to fork two pieces would be terrible, though).
|Dec-07-13|| ||Daisuki: <I don't think this would work work against good players>|
*I don't think this would work well against good players
|Dec-07-13|| ||SChesshevsky: This is interesting. I'm wondering if a reasonable test would be computer vs computer with a N less a pawn advantage, figuring if the weaker can draw a good amount maybe 50% or more of the time that might indicate that a computer, especially if its program could be tweaked, might have a good chance drawing a human most of the time even a N down.|
Maybe if a computer vs computer being down a n less 2 pawns draws virtually all the time, it might confirm the difficulty in beating a machine with less than full piece odds, which may end up happening often after the middlegame in match games.
I'm also wondering if B odds are more favorable to the weaker given the possibilities of blockading on color.
|Dec-07-13|| ||Daisuki: <SChesshevsky: This is interesting. I'm wondering if a reasonable test would be computer vs computer with a N less a pawn advantage, figuring if the weaker can draw a good amount maybe 50% or more of the time that might indicate that a computer, especially if its program could be tweaked, might have a good chance drawing a human most of the time even a N down.>|
Well, I certainly find it more difficult against Houdini if I blunder a pawn for no real compensation while up the knight. I think this would generally still be won for elite players, though, as a single pawn just doesn't accomplish too much in either the middlegame or the endgame so long as the extra piece exists. Two pawns becomes a lot more of a problem, including in endgames.
<Maybe if a computer vs computer being down a n less 2 pawns draws virtually all the time, it might confirm the difficulty in beating a machine with less than full piece odds, which may end up happening often after the middlegame in match games.>
Based on limited testing it seems like pawnstorming by the player down a knight and king safety for the player down multiple pawns (although I could change the pawns; I would take away a7, h7, b7, g7, then c7, depending on how far I go with that) both become problems for black. These are just 3'+3" games where I removed 3-5 pawns from black, though, and particularly in endgames Houdini doesn't seem to be playing that well (or perhaps I just am able to notice it more in endgames, which I'm better at).
I'm not sure how relevant this is to human vs. engine full knight odds situations, since 2+ pawns are worth a good deal, while 0 pawns is 0 pawns so long as there isn't a material blunder by the human. As I've said, it really seems like positional blunders that aren't rather serious and/or frequent are just magically made up by having the full extra knight. It seriously just gives an extra 0.1 pawns worth of position per turn. Probably more, actually, since even when the evaluation doesn't particularly change I'm usually not making the best moves.
I started a 150'+30" Houdini-Houdini game (Nb1, a7, h7 removed; ponder is off so it doesn't end up unfairly having one side think and the other side quickly reply for several moves in a row; it doesn't seem to actually run two separate Houdinis (and there's no real point in doing so as it would have to run both at the same time on both sides' moves, as opposed to one at a time on that side's move with ponder off), so it's unfair) for our interest, so I'll see how that goes. On move one white evaluated it at -2.08 (at depth 24, and it thought for 233 seconds; surprisingly low (it often blows 5% of its time per move), but I guess you do need about triple the total nodes per extra ply, so perhaps it didn't think seriously trying for depth 25 was worth it) and played d4, so it's up to black to get the win, I guess. This game should be finished in several hours.
<I'm also wondering if B odds are more favorable to the weaker given the possibilities of blockading on color.>
I would expect bishop odds to be better than knight odds because black alone starts with the bishop pair, and especially after trading queens is stronger on the color of the bishop removed for white (black for the queenside bishop). I guess it would help to avoid trading both of queens and the bishop of the color of white's bishop without getting white's bishop in the trading. If black traded both without getting white's bishop then white would indeed gain some advantage on the color of his unopposed bishop.
|Dec-08-13|| ||Daisuki: The game is finished.
Houdini 1.5a-Houdini 1.5a, 150'+30", 2013/12/07-2013/12/08, Nb1, a7, h7 removed:
1. d4 d5 2. Nf3 Nf6 3. Bf4 Ne4 4. e3 Bg4 5. Be2 e6 6. a3 Nc6 7. Nd2 Bxe2 8. Qxe2 Nxd2 9. Qxd2 Bb4 10. c3 Bd6 11. Bxd6 Qxd6 12. h3 Qe7 13. a4 Na5 14. Qd1 O-O-O 15. Qe2 Rd6 16. Rc1 e5 17. Kf1 Rb6 18. b4 Nc4 19. Re1 Kb8 20. b5 e4 21. Kg1 Qa3 22. Qc2 Rbh6 23. Rb1 f5 24. Qd1 Qxc3 25. Qe1 Qxe1+ 26. Rxe1 g5 27. Kf1 Nb2 28. Ra1 g4 29. Ke2 gxh3 30. g3 Nd3 31. Ra2 b6 32. Kf1 Kb7 33. Kg1 Ne1 34. Kf1 Nf3 35. Ra1 Ra8 36. Ke2 Ra5 37. Kd1 h2 38. Ke2 Rh8 39. Rad1 Rxa4 40. Kf1 Ra2 41. Kg2 Rha8 42. Rc1 R8a5 43. g4 Ra8 44. Kh3 Rg8 45. Rc6 Rxf2 46. Kg3 Ra2 47. Kf4 Rxg4+ 48. Kxf5 Rg1 49. Ke6 Rxh1 50. Kd7 Ra7 51. Kd8 Kb8 52. Rh6 Ng5 53. Kd7 Kb7 54. Rg6 Rf1 55. Rxg5 h1=Q 56. Ke6 Qh6+ 57. Kxd5 Qxg5+ 58. Kc4 Rc1+ 59. Kb4 Qe7+ 60. Kb3 Qa3# 0-1
With evaluations/depths and seconds per move (0s isn't always a literal 0.000s, as all time values are truncated):
1. d4 -2.08/24 233s d5 -2.26/24 226s
2. Nf3 -2.07/24 366s Nf6 -2.26/24 331s
3. Bf4 -2.07/23 153s Ne4 -2.29/23 284s
4. e3 -2.12/23 273s Bg4 -2.27/23 280s
5. Be2 -2.05/23 290s e6 -2.23/24 157s
6. a3 -2.15/23 158s Nc6 -2.33/23 319s
7. Nd2 -2.24/24 184s Bxe2 -2.47/25 146s
8. Qxe2 -2.30/25 146s Nxd2 -2.40/26 236s
9. Qxd2 -2.22/26 206s Bb4 -2.41/26 176s
10. c3 -2.75/25 0s Bd6 -2.40/27 447s
11. Bxd6 -2.29/26 174s Qxd6 -2.23/25 162s
12. h3 -2.12/24 158s Qe7 -2.20/23 198s
13. a4 -2.07/25 174s Na5 -2.33/24 132s
14. Qd1 -2.23/23 221s O-O-O -2.46/23 119s
15. Qe2 -2.28/23 370s Rd6 -2.46/23 209s
16. Rc1 -2.58/23 209s e5 -3.03/25 889s
17. Kf1 -3.05/24 333s Rb6 -3.07/23 106s
18. b4 -3.16/24 678s Nc4 -2.86/23 0s
19. Re1 -3.20/24 104s Kb8 -3.75/23 344s
20. b5 -3.59/23 295s e4 -3.96/25 105s
21. Kg1 -3.84/25 101s Qa3 -4.35/24 124s
22. Qc2 -4.27/24 141s Rbh6 -4.66/25 349s
23. Rb1 -4.65/24 207s f5 -4.35/23 0s
24. Qd1 -4.87/23 364s Qxc3 -4.57/22 0s
25. Qe1 -5.16/24 235s Qxe1+ -4.88/23 0s
26. Rxe1 -4.78/22 17s g5 -4.57/24 91s
27. Kf1 -4.73/24 115s Nb2 -4.56/23 0s
28. Ra1 -4.76/24 203s g4 -4.59/23 0s
29. Ke2 -4.92/25 195s gxh3 -5.39/26 210s
30. g3 -5.59/24 219s Nd3 -5.44/23 0s
31. Ra2 -5.41/25 164s b6 -5.26/24 0s
32. Kf1 -5.47/26 116s Kb7 -5.32/25 0s
33. Kg1 -6.13/28 523s Ne1 -5.31/26 0s
34. Kf1 -7.39/22 67s Nf3 -7.24/21 0s
35. Ra1 -7.44/23 74s Ra8 -7.30/22 0s
36. Ke2 -9.96/27 264s Ra5 -9.82/26 0s
37. Kd1 -9.83/26 82s h2 -9.68/25 0s
38. Ke2 -10.43/25 67s Rh8 -10.28/24 0s
39. Rad1 -11.11/31 373s Rxa4 -12.38/27 48s
40. Kf1 -12.65/29 87s Ra2 -12.51/28 0s
41. Kg2 -13.51/29 306s Rha8 -12.84/27 0s
42. Rc1 -14.53/26 312s R8a5 -14.40/25 0s
43. g4 -15.17/26 165s Ra8 -15.04/25 0s
44. Kh3 -23.32/25 229s Rg8 -16.94/23 0s
45. Rc6 -136.43/24 110s Rxf2 -M33/23 23s
46. Kg3 -M22/25 80s Ra2 -M53/24 12s
47. Kf4 -136.49/25 166s Rxg4+ -M21/24 38s
48. Kxf5 -M20/1 0s Rg1 -19.35/8 0s
49. Ke6 -M13/25 52s Rxh1 -M13/24 0s
50. Kd7 -M11/27 46s Ra7 -M11/26 1s
51. Kd8 -M10/28 79s Kb8 -M10/27 1s
52. Rh6 -M9/28 86s Ng5 -M9/27 1s
53. Kd7 -M8/26 2s Kb7 -M8/25 0s
54. Rg6 -M7/24 8s Rf1 -M7/23 0s
55. Rxg5 -M6/22 0s h1=Q -M6/21 0s
56. Ke6 -M5/20 0s Qh6+ -M5/19 0s
57. Kxd5 -M4/18 0s Qxg5+ -M4/17 0s
58. Kc4 -M3/16 0s Rc1+ -M3/15 0s
59. Kb4 -M2/14 0s Qe7+ -M2/13 0s
60. Kb3 -M1/1 0s Qa3# -M1/1 0s
Apparently I can't make it so black refuses to use white's evaluation or vice versa (it was almost always simply black satisfied enough with white's huge evaluation in its favor to spend none of its time thinking that move), so I could use another engine (maybe a better one) for white and see what happens.
|Dec-08-13|| ||SChesshevsky: <Daisuki: The game is finished.>|
Thanks for your time and effort. I'm going to play through this game and try to get a hands on feel for the programs "thinking".
I also posted here:
Stockfish vs Houdini, 2013
Do you know what version Houdini was played in that game?
|Dec-08-13|| ||Daisuki: I did a 3'+2" tournament between Houdini 1.5a, Stockfish DD (not sure what the version number should be, but it's the latest one on their website), Komodo 3, and Rybka 2.2 where Nb1, a7, and h7 were missing in each game. Black won ten games and the other two were draws. 11.0/12.0 is a +417 or so performance. However, the only draws were in games where Rybka was black against Houdini and Stockfish, respectively. On the CCRL list Rybka 2.2 is 2982, Komodo 3 is 3108, Houdini 1.5a is 3201, and Stockfish 4 is 3219. So even Stockfish is only 237 points above Rybka while Houdini is 219 above and Komodo is 126 points above. If you take out the "417" for having black then Rybka is 180, 198, and 291 points above the others, respectively. This corresponds to expected scores of about 74%, 76%, and 84%, respectively. Which would mean 48%, 52%, and 68% win rates if the other results were always draws (and higher win rates if losses happened). Rybka failed to win either of its "50/50" games, while it won the 68% game. 1/3 wins as opposed to 1.68/3 wins "expected". Komodo is also the clearly inferior engine of the three Rybka faced and was the only one Rybka beat with black. This is quite possibly just randomness due to the fast time control, but it's interesting that Rybka underperformed, while Komodo, clearly third on CCRL (of the engines I used), was clearly third in the tournament as well. Among the top three engines black scored 100%.|
<SChesshevsky: <Daisuki: The game is finished.>
Thanks for your time and effort. I'm going to play through this game and try to get a hands on feel for the programs "thinking".
I also posted here:
Stockfish vs Houdini, 2013
Do you know what version Houdini was played in that game?>
The game is #328 of nTCEC Season 1, which is listed as Stockfish 250413 vs. Houdini 3.
|Dec-08-13|| ||SChesshevsky: < Daisuki: The game is finished.
Houdini 1.5a-Houdini 1.5a, 150'+30", 2013/12/07-2013/12/08, Nb1, a7, h7 removed:>|
Thanks for the game. I only had time to run through it once but here are my impressions.
I'm more convinced that for a machine to play optimally a piece down less pawn or pawns, the program would have to be tweaked.
In this game, it looked like White played "straight" chess but being hurt by his advantage of a & h pawns, which turned out to be a disadvantage because of the open files, was not inclined to castle and connect his Rooks quickly.
It also looked like, not being able to plan, White didn't consider that to get equality it'd probably necessitate getting an outside passed pawn and work under that basis as soon as possible without distraction. For instance White's probably best served by moves after 14...0-0-0 being focused on gaining space on the Kside, mobilizing forces there and pushing those pawns while trying to minimize tempo loss. The Kside push also facilitates the exchange of pawns, which as we discussed, the more pawns off the board the more the value of the extra piece is reduced.
I guess the computer played the best moves for White but it sure didn't feel that way.
|Dec-09-13|| ||kardopov: <shach matov: There is no need to try anything, I know that playing against a commercial program on you PC is very very different from playing against a special program in a match. I played many computers, I know what that means, it's not going to change my mind about the match>|
It's a no-brainer. Just let the two combatants play against each other and presto!, you'll get the answer. The real problem is, who will sponsor the match to make it happen?
|Dec-09-13|| ||kardopov: Gee! Knight handicap for a super powerful computer against the strongest human chess player of today. That would be very interesting. It's whetting my appetite. I bet Carlsen will smash the super computer to oblivion.|
|Dec-09-13|| ||shach matov: kardropov - I know, the interest will be huge!! People are still talking about it on the site. I think the machine will be the favorite, but it could be close! Too bad nobody seems to be interested in doing such a match, which is a shame, could've been be much fun|
|Dec-09-13|| ||Marmot PFL: Of course nobody with any sense is interested in such a ridiculous match (including you I actually think...) The hardest part would be to keep the computer from resigning.|
|Dec-09-13|| ||Daisuki: <SChesshevsky>, it's true that it doesn't "plan" in that sense, although as you could see it did look ahead about a dozen moves (for each side) more or less all the time. I think that to us it may look weirder because we underestimate the problems that could be faced if we simply castled into black's half-open h file manned by the Rh8. If it was just me in a game with someone on my level then it would be much easier, but computers are ruthless at exploiting both a lack of king safety and passive positions, so even if you fix the former by causing the latter it can be a problem.|
Anyway, pushing pawns aggressively also has its drawbacks, namely how it becomes easier to attack the pushed pawns. It is still down a piece, so it may be difficult to do this without losing pawns or making other major concessions that lead to lost material or just a loss in the future. This seemed to be something white could handle better in the games where it had 3-5 extra pawns for the knight, but in the longer time control game where it had 2 extra pawns it obviously didn't see a good way to get enough control of the board to make much of what it considers progress. Another thing to keep in mind is that even if it gets a passed pawn it likely can't promote it while down a piece (and therefore doesn't exactly jump at that plan unless it really thinks it looks decent; it does of course value passed pawns, but I doubt it does if it doesn't expect to hold and make use of them). And getting two clear, solid passed pawns without compromising king safety seems like a tall order while down a piece.
I'll try a Stockfish-Houdini 150'+30" game missing Nb1, a7, and h7, where Stockfish is set with -50 contempt (and should thus jump at draws), 200 aggressiveness, and 0 cowardice. These are the minimum, maximum, and minimum numbers, respectively, the defaults are 0, 100, and 100, respectively, and 50, 0, and 200 are the maximum, minimum, and maximum numbers, respectively. I haven't looked into what these settings do well, so I don't know how much they might help or hurt. ;p
|Dec-09-13|| ||shach matov: Marmont - I would personally be very interested (do i have a reason to lie?) and many people already expressed their interest and even kardropov now also shared this interest. The best engine (not something we can buy but specifically prepared for the match) with the most powerful hardware should win, but it may be close. But there is no doubt that the interest will be huge!! It would be a fascinating thing!|
|Dec-09-13|| ||Daisuki: I'm canceling this Stockfish-Houdini game, as Stockfish is performing poorly. I'll try the other three extreme combinations for aggressiveness and cowardice in 3'+2" games to try to find a better combination for a long game.|
|Dec-09-13|| ||Daisuki: I think 0/0 aggressiveness/cowardice might be best, as it took much longer to get mated because it pushed many pawns and didn't trade pieces quickly. I tested all four extreme combinations and the 100/100 default combination, all with -50 contempt. I don't know when I'll try a long game.|
|Dec-09-13|| ||Marmot PFL: Looks at it from Carlsen's POV. If he wins, big deal he has a big edge to begin with, and if he loses it would be too embarrassing.|
|Dec-10-13|| ||SChesshevsky: <Daisuki: ...if we simply castled into black's half-open h file manned by the Rh8.... Anyway, pushing pawns aggressively also has its drawbacks..>|
Related to castling, that's my point the pawn's advantage actually turned into a disadvantage because the King would not castle. As in the game, the King moves 7 times by move 35 all on the back rank. That doesn't exactly suggest proper play needed to achieve equality.
Related to pawn pushing, I'm obviously not advocating woodpusher pawn attacks but with the Knight on the a-file, O-O-O, and if the Rooks were connected without tempo loss it looks like White has at least a shot at equality. You don't have to promote the pawn but gain space on the Kside, exchange the lead for another then try to keep the pressure on that minority side and look to keep exchanging pawns. Certainly not the same strategy as one would use to win even-up.
That's why I think a chess computer's program would likely have to be altered to favor lines which have the best chances for drawing rather than trying to gain any sort of advantage.
That's the idea of computer vs. computer. If the disadvantaged computer can draw a significant number of times against another similar computer, and if it can be tweaked to even be better at drawing, then the assumption is that against a top player with odds the computer might not win but might also rarely lose.
|Dec-10-13|| ||Daisuki: In the game black doubled rooks on the h file. So castling kingside probably would've been bad. White could've castled queenside after black did, but by then white had disturbed its pawn structure there. I think overall it's a problem because you can't just not disturb your pawn structure anywhere, so wherever you castle black can just double rooks (or set up Alekhine's gun) and slam you. The more you trade pawns, the more mobility that extra piece gets, too, which is also a problem. Even if you play better than your opponent and get a passed pawn for no material or positional concession I think that passed pawn would just get eaten later. And I really don't see how to get multiple passed pawns without issues. Honestly I would expect a more passive strategy based on having pawns in front and trading and opening the position as little as possible to perhaps have better chances, but when I personally try this while up the full knight against Houdini or Stockfish it's a fine line between holding things together and getting so passive that I fall apart at least a little. Engines could do better, but how much better I don't know.|
|Dec-11-13|| ||SChesshevsky: <Daisuki: ... a more passive strategy based on having pawns in front and trading and opening the position as little as possible to perhaps have better chances...>|
A great discussion. Lots to think about. Related to the above comment and unrelated to computer play, the general theory is that with a material disadvantage, even a pawn, passive play typically loses outright. The strategy with advantage is to get good position, easier when there are no complications, then position the King and then exchange for the endgame. It's easier with the Rooks off. Best defense is usually to try to create an unbalance somewhere, stir up complications and swindle.
It's probably ancient history now but when I played a lot, being a pawn up, especially the right one, meant a won game. Byrne once wrote about Fischer's poisoned pawn defense that though risky, the pawn if held means a won game. Or something like that. It might be something to consider for own play as it might still be true.
|Dec-11-13|| ||kardopov: C'mmon, start the nudging. Who's gonna hurl the first salvo? Maybe a program developer should challenge Carlsen to a winner take all match with the prize coming from both parties as a wager, say to the amount of 2M Dollars. Both parties have to raise 1M each. The condition: 7 games of classical chess, less knight for the computer.|
|Dec-11-13|| ||Daisuki: I know passive play is generally bad, but engines default to active play, and that seems like a loser when you're down a piece, unless you have <many> extra pawns. Two pawns doesn't seem to be very close to cutting it (in terms of expected score; I know the ~200-point inferior Rybka couldn't win in a couple of its 3'+2" games, but more equal engines always did so far) between two roughly equal strong engines. "Swindles" just aren't really going to happen in such engine games, as "complications" are no problem for engines, and, as stated by others, tactics favor the player with the extra piece. I'm sure it would work better between more unequal players and/or players enough lower in skill so as to be unable to exploit the extra piece too well relative to the two extra pawns. Otherwise complications and active play are like marching forward to your death, since you are down the piece. Of course everything's a matter of death, but it did take a third more moves in a 3'+2" game for Houdini to mate Stockfish when Stockfish was set to 0/0 aggressiveness/cowardice (which seems to mean that it doesn't care much about both reducing the opponent's king safety and increasing its own king safety, respectively). Under this setting it didn't seem to place its pieces so aggressively or trade them much, and it kept the position more closed by pushing pawns, which seems better than normal in this situation.|
Yes, I'd say that an extra pawn from a starting position would probably often be won (between equal enough, good enough players).
|Dec-11-13|| ||Daisuki: -50/0/0 contempt factor/aggressiveness/cowardice Stockfish lost to Houdini in 56 moves, while -50/100/100 (the latter two being defaults) lost in 55 moves. Both games were 60'+30". I don't really see an easy way to make it drag out the game. I could play with pawn settings, but in the end the piece is worth way more than the two pawns to engines. The starting position with Nb1, a7, and h7 missing is evaluated at -2 to -3 by various engines, in spite of white having the advantage of the first move.|
< Earlier Kibitzing · PAGE 7 OF 7 ·