Author: James Robertson
Date: 13:15:46 09/20/99
Go up one level in this thread
On September 20, 1999 at 14:51:35, Robert Hyatt wrote: >On September 20, 1999 at 14:08:26, Jon Dart wrote: > >>Arasan doesn't generate rook or bishop underpromotions except at >>ply 0 (I think Crafty does the same thing). > > >No. Crafty generates all promotions in normal search, but only generates >queen promotions in the q-search. I think leaving out underpromotions is a >mistake in the regular search... but in the q-search only doing queens seems >reasonable if you don't handle checks and escaping from check. If you do, I'd >likely throw in knight promotions... > > My program only generates Queen and Knight promotions, as I have never seen a position where a Bishop/Rook promotion was any good.... I have run across a few nice Knight promotion positions, though. James > > >> >>Forward pruning is a reasonable thing to do. In the qsearch, I >>do something like: >> >>if (current material + gain from promotion + some constant < alpha) >> skip move >> >>My latest build has this output on your example: >> >>ply 4. Kh3 0 seconds. score: -7.19 215 nodes. >>ply 5. Kh3 0 seconds. score: -7.19 283 nodes. >>ply 5. Kh3 0 seconds. score: -11.08 470 nodes. >>ply 6. Kh3 0 seconds. score: -11.08 1144 nodes. >>ply 7. Kh3 0 seconds. score: -11.11 2538 nodes. >>ply 8. Kh3 0 seconds. score: -11.28 4856 nodes. >>ply 9. Kh3 0 seconds. score: -11.28 5172 nodes. >>ply 9. Kh3 0 seconds. score: -17.02 9453 nodes. >>ply 10. Kh3 0 seconds. score: -17.52 19403 nodes. >>ply 11. Kh3 0 seconds. score: -17.52 20700 nodes. >>ply 11. Kh3 0 seconds. score: -156.00 50050 nodes. >> >>--Jon >> >>On September 20, 1999 at 02:09:30, Steffen Jakob wrote: >> >>>Hi all! >> >>> >>>I still try to improve the playing skill of Hossa in pawn endgames (which is not >>>very difficult because he plays them very badly). About underpromotions: I know >>>that Amir Ban doesn't generate underpromotions in the search tree. I don't like >>>that very much. What I am doing now is to reduce the search depth for >>>underpromotions (one ply). It helps a lot in some positions, e.g. in the pawn >>>endgame position where crafty saw that mate in 4 although it is a mate in 6. As >>>you can see below I need 158616 nodes to see the mate without pruning and 135713 >>>with underpromotion pruning. I think about reducing the depth by two plies for >>>bishop underpromotions (can someone please give me some positions where the >>>winner move is to promote the pawn to a bishop?). What are you thinking about >>>this? Are there other ways to prune underpromotions? E.g. if you are sure that >>>queen promotions are searched before underpromotions and alpha > xxx then it >>>might be useless to try those moves (at least rook and bishop promotions). >>> >>>analysis without forward pruning: >>>--------------------------------- >>> >>> +---+---+---+---+---+---+---+---+ >>>8 | | | | | | | | | >>> +---+---+---+---+---+---+---+---+ Black: Hossa v1.63 (C) (0) >>>7 ||P|| | | | | | | | >>> +---+---+---+---+---+---+---+---+ Black material: 4 >>>6 | P | | | | | ||P|| | >>> +---+---+---+---+---+---+---+---+ >>>5 | | | | | | | ||P|| >>> +---+---+---+---+---+---+---+---+ White to move >>>4 | | | | | | | | P | >>> +---+---+---+---+---+---+---+---+ 8/p7/P5p1/7p/7P/4kpK1/8/8 w >>>3 | | | | ||K|||P|| K | | >>> +---+---+---+---+---+---+---+---+ White material: 2 >>>2 | | | | | | | | | >>> +---+---+---+---+---+---+---+---+ White: opponent (0) >>>1 | | | | | | | | | >>> +---+---+---+---+---+---+---+---+ >>> a b c d e f g h >>> >>> >>> depth value time nodes pv >>> 1 -5.23 0.00 2 1.g3h3 >>> <1> -5.23 0.00 3 1.g3h3 >>> 2 -5.35 0.00 38 1.g3h3 f2 2.h3g2 >>> <2> -5.35 0.00 48 1.g3h3 f2 2.h3g2 >>> 3 -5.37 0.00 84 1.g3h3 f2 2.h3g2 e3e2 >>> <3> -5.37 0.00 90 1.g3h3 f2 2.h3g2 e3e2 >>>? 4 -510.92 0.01 194 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>>? <4> -510.92 0.01 207 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>> 5 -510.92 0.01 322 1.g3h3 f2 2.h3g2 >>> <5> -510.92 0.01 335 1.g3h3 f2 2.h3g2 >>> 6 -510.92 0.01 531 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>> <6> -510.92 0.01 552 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>> 7 -510.92 0.02 1182 1.g3h3 f2 2.h3g2 e3e2 3.g2h2 f2f1Q >>> 4.h2g3 >>> <7> -510.92 0.02 1203 1.g3h3 f2 2.h3g2 e3e2 3.g2h2 f2f1Q >>> 4.h2g3 >>> 8 -510.98 0.05 2817 1.g3h3 f2 2.h3g2 e3e2 3.g2h2 e2f3 >>> 4.h2h3 f2f1Q 5.h3h2 >>> <8> -510.98 0.05 2867 1.g3h3 f2 2.h3g2 e3e2 3.g2h2 e2f3 >>> 4.h2h3 f2f1Q 5.h3h2 >>> 9 -510.97 0.09 5587 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>> 4.g3h2 f1f2 5.h2h3 e2f3 >>> <9> -510.97 0.09 5644 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>> 4.g3h2 f1f2 5.h2h3 e2f3 >>>? 10 -512.01 0.24 14405 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>> 4.g3h2 e2e3 5.h2g3 f1xa6 >>>? <10> -512.01 0.24 14556 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>> 4.g3h2 e2e3 5.h2g3 f1xa6 >>> 11 -512.03 0.64 41493 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>> 4.g3h2 e2e3 5.h2g3 f1xa6 6.g3g2 >>> <11> -512.03 0.65 41840 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>> 4.g3h2 e2e3 5.h2g3 f1xa6 6.g3g2 >>>? 12 #-6 2.11 158616 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>> 4.g3h2 g5 5.h4xg5 e2f3 6.g6 f1g2 >>>? <12> #-6 2.12 159367 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>> 4.g3h2 g5 5.h4xg5 e2f3 6.g6 f1g2 >>> >>> >>>analysis with forward pruning: >>>------------------------------ >>> >>> depth value time nodes pv >>> 1 -5.23 0.00 2 1.g3h3 >>> <1> -5.23 0.00 3 1.g3h3 >>> 2 -5.35 0.00 38 1.g3h3 f2 2.h3g2 >>> <2> -5.35 0.00 48 1.g3h3 f2 2.h3g2 >>> 3 -5.37 0.00 84 1.g3h3 f2 2.h3g2 e3e2 >>> <3> -5.37 0.00 90 1.g3h3 f2 2.h3g2 e3e2 >>>? 4 -510.92 0.01 194 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>>? <4> -510.92 0.01 207 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>> 5 -510.92 0.02 322 1.g3h3 f2 2.h3g2 >>> <5> -510.92 0.02 335 1.g3h3 f2 2.h3g2 >>> 6 -510.92 0.03 531 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>> <6> -510.92 0.03 552 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>> 7 -510.92 0.03 1182 1.g3h3 f2 2.h3g2 e3e2 3.g2h2 f2f1Q >>> 4.h2g3 >>> <7> -510.92 0.03 1203 1.g3h3 f2 2.h3g2 e3e2 3.g2h2 f2f1Q >>> 4.h2g3 >>> 8 -510.98 0.04 2726 1.g3h3 f2 2.h3g2 e3e2 3.g2h2 e2f3 >>> 4.h2h3 f2f1Q 5.h3h2 >>> <8> -510.98 0.04 2776 1.g3h3 f2 2.h3g2 e3e2 3.g2h2 e2f3 >>> 4.h2h3 f2f1Q 5.h3h2 >>> 9 -510.97 0.08 5403 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>> 4.g3h2 f1f2 5.h2h3 e2f3 >>> <9> -510.97 0.08 5460 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>> 4.g3h2 f1f2 5.h2h3 e2f3 >>>? 10 -512.01 0.16 12077 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>> 4.g3h2 e2e3 5.h2g3 f1xa6 >>>? <10> -512.01 0.16 12228 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>> 4.g3h2 e2e3 5.h2g3 f1xa6 >>> 11 -512.03 0.46 33004 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>> 4.g3h2 e2e3 5.h2g3 f1xa6 6.g3g2 >>> <11> -512.03 0.46 33351 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>> 4.g3h2 e2e3 5.h2g3 f1xa6 6.g3g2 >>>? 12 #-6 1.71 135713 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>> 4.g3h2 g5 5.h4xg5 e2f3 6.g6 f1g2 >>>? <12> #-6 1.72 136462 1.g3h3 f2 2.h3g2 e3e2 3.g2g3 f2f1Q >>> 4.g3h2 g5 5.h4xg5 e2f3 6.g6 f1g2 >>> >>> >>>Best wishes, >>>Steffen.
This page took 0.01 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.