Author: Henk Bossinade
Date: 08:33:04 08/14/04
Go up one level in this thread
On August 14, 2004 at 09:19:59, Stuart Cracraft wrote: >On August 11, 2004 at 23:36:53, Robert Hyatt wrote: > >>On August 11, 2004 at 17:22:04, Stuart Cracraft wrote: >> >>>On August 10, 2004 at 13:07:21, Robert Hyatt wrote: >>> >>>>On August 09, 2004 at 14:26:18, Stuart Cracraft wrote: >>>> >>>>>Sorry, the posted position was in error. It is: >>>>> >>>>>>[D]r1bqkb1r/2pp1ppp/p1n2n2/1p2p3/4P3/1B3N2/PPPP1PPP/RNBQ1RK1 b q - 4 1 >>>>> >>>>>On August 09, 2004 at 14:25:30, Stuart Cracraft wrote: >>>>> >>>>>>[D]r1bqkb1r/2pp1ppp/p1n2p2/1p2p3/4P3/1B3N2/PPPP1PPP/RNBQ1RK1 b q - 4 1 >>>>>> >>>>>>I don't know if I filled in the castle flags right but in this position >>>>>>black still has both rights left. >>>>>> >>>>>>The question is, if your program values bishops more than knights, >>>>>>how do you prevent it from playing Na5 to capture the White Bishop >>>>>>but then giving up the pawn at e5 (since the capture Nxe4 by black >>>>>>is unplayable). (e.g. 1. ... Na5 2. Nxe5 Nxb3 3. ab) >>>>>> >>>>>>I've searched this to 10 full ply with unlimited quiescence >>>>>>and it still can't see that Na5 is bad. >>>>>> >>>>>>It would help if some people posted their searches and mention >>>>>>how they avoid this kind of issue. >>>>>> >>>>>>My search: >>>>>> >>>>>>Alpha=-255 Beta=245 Maxdepth=99 MaxTime=9999999 >>>>>> 1/11> c6a5 0.00 245 68 c6a5 >>>>>> 1/11 f6e4 0.01 1026 159 f6e4 >>>>>> 2/15 f6e4 0.01 1008 946 f6e4 f1e1 >>>>>> 3/17< f6e4 0.02 758 4181 f6e4 d2d4 e5d4 >>>>>> 3/20 f6e4 0.04 75 7143 f6e4 b3d5 e4f6 >>>>>> 4/30 f6e4 0.13 75 24970 f6e4 b3d5 e4f6 d5c6 >>>>>> 5/34 c6a5 0.50 57 104715 c6a5 f1e1 a5b3 a2b3 c8b7 >>>>>> 6/34 c6a5 1.22 45 256985 c6a5 f1e1 a5b3 a2b3 c8b7 b1c3 >>>>>> 7/42 c6a5 4.31 32 889215 c6a5 f3e5 a5b3 c2b3 f8d6 d2d4 f6e4 >>>>>> 8/51 c6a5 24.86 19 5021946 c6a5 f3e5 a5b3 c2b3 d8e7 d2d4 d7d6 e5f3 >>>>>> 9/54 c6a5 115.27 19 19483875 c6a5 f3e5 a5b3 c2b3 d8e7 d2d4 d7d6 e5f3 >>>>>>10/54 c6a5 264.28 26 44999682 c6a5 f1e1 a5b3 a2b3 c8b7 d2d4 f6e4 b1d2 d7d5 >>>>> >>>>>> >>>>>>Stuart >>>> >>>> >>>>The output above looks broken. Why not just take the pawn. Here is Crafty >>>>after I play Na5 as black: >>>> >>>> 1 0.09 0.47 2. Nxe5 Nxb3 3. axb3 Nxe4 >>>> 1-> 0.09 0.47 2. Nxe5 Nxb3 3. axb3 Nxe4 >>>> 2 0.09 0.47 2. Nxe5 Nxb3 3. axb3 Nxe4 >>>> 2-> 0.09 0.47 2. Nxe5 Nxb3 3. axb3 Nxe4 >>>> 3 0.09 0.47 2. Nxe5 Nxb3 3. axb3 Nxe4 >>>> 3-> 0.09 0.47 2. Nxe5 Nxb3 3. axb3 Nxe4 >>>> 4 0.10 0.61 2. Nxe5 Nxb3 3. axb3 Bd6 4. d4 Bxe5 >>>> 5. dxe5 Nxe4 >>>> 4-> 0.10 0.61 2. Nxe5 Nxb3 3. axb3 Bd6 4. d4 Bxe5 >>>> 5. dxe5 Nxe4 >>>> 5 0.10 0.61 2. Nxe5 Nxb3 3. axb3 Bd6 4. d4 Bxe5 >>>> 5. dxe5 Nxe4 >>>> 5-> 0.10 0.61 2. Nxe5 Nxb3 3. axb3 Bd6 4. d4 Bxe5 >>>> 5. dxe5 Nxe4 (s=2) >>>> 6 0.10 0.61 2. Nxe5 Nxb3 3. axb3 Bd6 4. d4 Bxe5 >>>> 5. dxe5 Nxe4 >>>> 6-> 0.11 0.61 2. Nxe5 Nxb3 3. axb3 Bd6 4. d4 Bxe5 >>>> 5. dxe5 Nxe4 (s=2) >>>> 7 0.11 0.80 2. Nxe5 Nxb3 3. axb3 Bd6 4. d4 Bb7 >>>> 5. f3 O-O >>>> 7-> 0.12 0.80 2. Nxe5 Nxb3 3. axb3 Bd6 4. d4 Bb7 >>>> 5. f3 O-O (s=2) >>>> 8 0.15 0.85 2. Nxe5 Nxb3 3. axb3 d6 4. Nf3 Be6 >>>> 5. e5 Ne4 6. d4 >>>> 8-> 0.17 0.85 2. Nxe5 Nxb3 3. axb3 d6 4. Nf3 Be6 >>>> 5. e5 Ne4 6. d4 >>>> 9 0.22 0.65 2. Nxe5 Nxb3 3. axb3 d6 4. Nc6 Qd7 >>>> 5. Nb4 Nxe4 6. d3 c5 >>>> 9-> 0.35 0.65 2. Nxe5 Nxb3 3. axb3 d6 4. Nc6 Qd7 >>>> 5. Nb4 Nxe4 6. d3 c5 (s=3) >>>> 10 0.44 0.97 2. Nxe5 Nxb3 3. axb3 d6 4. Nc6 Qd7 >>>> 5. Nb4 Be7 6. d4 O-O 7. Nc3 (s=2) >>>> 10-> 0.54 0.97 2. Nxe5 Nxb3 3. axb3 d6 4. Nc6 Qd7 >>>> 5. Nb4 Be7 6. d4 O-O 7. Nc3 (s=2) >>>> 11 0.70 0.78 2. Nxe5 Nxb3 3. axb3 d6 4. Nc6 Qd7 >>>> 5. Nb4 Be7 6. d4 O-O 7. Nc3 Bb7 >>>> 11-> 2.44 0.78 2. Nxe5 Nxb3 3. axb3 d6 4. Nc6 Qd7 >>>> 5. Nb4 Be7 6. d4 O-O 7. Nc3 Bb7 (s=4) >>>> 12 3.03 0.83 2. Nxe5 Nxb3 3. axb3 d6 4. Nc6 Qd7 >>>> 5. Nd4 Be7 6. Qf3 O-O 7. Nc3 b4 8. >>>> e5 (s=3) >>>> 12-> 6.44 0.83 2. Nxe5 Nxb3 3. axb3 d6 4. Nc6 Qd7 >>>> 5. Nd4 Be7 6. Qf3 O-O 7. Nc3 b4 8. >>>> e5 (s=5) >>>> 13 10.00 0.88 2. Nxe5 Nxb3 3. axb3 d6 4. Nc6 Qd7 >>>> 5. Nb4 Be7 6. Nc3 O-O 7. Qf3 Bb7 8. >>>> d4 Qg4 <HT> (s=4) >>>> 13-> 18.39 0.88 2. Nxe5 Nxb3 3. axb3 d6 4. Nc6 Qd7 >>>> 5. Nb4 Be7 6. Nc3 O-O 7. Qf3 Bb7 8. >>>> d4 Qg4 <HT> (s=4) >>>> 14 31.39 0.82 2. Nxe5 Nxb3 3. axb3 Bb7 4. d4 Be7 >>>> 5. f3 d6 6. Nd3 O-O 7. c4 c5 8. cxb5 >>>> axb5 9. Rxa8 Bxa8 (s=3) >>>> 14-> 55.88 0.82 2. Nxe5 Nxb3 3. axb3 Bb7 4. d4 Be7 >>>> 5. f3 d6 6. Nd3 O-O 7. c4 c5 8. cxb5 >>>> axb5 9. Rxa8 Bxa8 (s=3) >>>> 15 1:28 0.80 2. Nxe5 Nxb3 3. axb3 Be7 4. d4 O-O >>>> 5. f3 d6 6. Nc6 Qd7 7. Nxe7+ Qxe7 8. >>>> Bg5 h6 9. Bxf6 Qxf6 10. Nc3 Re8 (s=2) >>>> 15-> 3:00 0.80 2. Nxe5 Nxb3 3. axb3 Be7 4. d4 O-O >>>> 5. f3 d6 6. Nc6 Qd7 7. Nxe7+ Qxe7 8. >>>> Bg5 h6 9. Bxf6 Qxf6 10. Nc3 Re8 (s=4) >>> >>>Black cannot play Na5 because of Nxe5 and the loss of a pawn. >>> >>>So my question was / still is, what should Black instead of Na5 >>>if the program values Knights more than Bishops? Mine plays Na5, >>>gets the Bishop for the Knight and then promptly loses a center >>>pawn. I could decrease the value of the Bishop or increase the >>>value of the center pawn, but both of these seem a little extreme >>>given Bishop vs. Knight and the current high value of a center pawn. >>>The point is that a center pawn in this program is not worth as much >>>as a Bishop minus a Knight. The former is maybe 30 or 40 millipawns >>>and the latter is maybe 100 or 200 millipawns. I am looking for good >>>ways to avoid 1. ... Na5 in the first place. >>> >>>A knight on the rim penalty doesn't do it since the Knight is just >>>traded off. >>> >>>Valuing the center pawn too much makes no sense. >>> >>>I am running out of ideas to prevent 1. ... Na5. >>> >>>Stuart >> >> >>Why would the program "value knights more than bishops"?? firstly??? >> >>Next, chess is a complex game that requires harmonious coordination between the >>various eval terms. Make one too big and it fails to work as planned. Or it >>causes other terms to appear to be too small, and making them bigger only makes >>things get worse. >> >>First, here is crafty in the position where you are wanting to play the knight >>to a5: >> >> 10 0.71 1.00 1. ... d6 2. Ng5 d5 3. exd5 Nd4 4. >> Re1 Ng4 5. Ne6 fxe6 6. Qxg4 Nxb3 <HT> >> 10-> 0.81 1.00 1. ... d6 2. Ng5 d5 3. exd5 Nd4 4. >> Re1 Ng4 5. Ne6 fxe6 6. Qxg4 Nxb3 <HT> >> 11 0.81 1.00 1. ... d6 2. Ng5 d5 3. exd5 Nd4 4. >> Re1 Ng4 5. Ne6 fxe6 6. Qxg4 Nxb3 <HT> >> 11-> 1.26 1.00 1. ... d6 2. Ng5 d5 3. exd5 Nd4 4. >> Re1 Ng4 5. Ne6 fxe6 6. Qxg4 Nxb3 <HT> >> 12 4.12 1.10 1. ... d6 2. Ng5 Be6 3. Bxe6 fxe6 4. >> Nxe6 Qd7 5. Nxf8 Rxf8 6. d3 Qf7 7. >> Nc3 Nd4 8. Bg5 >> 12-> 19.14 1.10 1. ... d6 2. Ng5 Be6 3. Bxe6 fxe6 4. >> Nxe6 Qd7 5. Nxf8 Rxf8 6. d3 Qf7 7. >> Nc3 Nd4 8. Bg5 (s=6) >> 13 30.75 1.04 1. ... d6 2. Ng5 Be6 3. Bxe6 fxe6 4. >> Nxe6 Qd7 5. Nxf8 Rxf8 6. d3 Qg4 7. >> f3 Qh4 8. Be3 d5 9. exd5 Nxd5 (s=5) >> 13-> 50.84 1.04 1. ... d6 2. Ng5 Be6 3. Bxe6 fxe6 4. >> Nxe6 Qd7 5. Nxf8 Rxf8 6. d3 Qg4 7. >> f3 Qh4 8. Be3 d5 9. exd5 Nxd5 >> 14 1:17 1.23 1. ... d6 2. Ng5 d5 3. exd5 Nd4 4. >> d6 Nxb3 5. dxc7 Qxc7 6. cxb3 Bg4 7. >> f3 Bf5 8. Re1 Rd8 9. Qe2 >> >> >> >>Black has already played some pawn moves that leave white looking fine. Why >>move the knight two more times, trading three tempi for the white bishop? Black >>needs to develop. Develop. Develop. Not waste time chasing an unimportant >>bishop. Later, maybe. But if black wastes too much time, the game will slip >>away. The first few moves are not about bishops vs knights, because it is not >>yet clear which will end up being more valuable. The opening is to develop >>pieces first. _then_ start with other things... IE A GM will tell you that >>three tempi are worth a pawn. Three tempi for that bishop are hardly worth that >>cost... > >I realize and understand this but wonder what is the best method >to avoid a Na5. > >Penalize repeat opening moves by the same piece? >Have large rim penalties for knights? >Value bishops equally to bishops in the opening? >I'm sure it's a long list. I think you should simply give your engine more time. Crafty 19.1 with material eval only and BISHOP_VALUE changed from 300 to 330 drops Na5 and turns back to Nxe4 at depth 11. The same goes for my own engine.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.