Author: leonid
Date: 10:12:35 05/06/01
Go up one level in this thread
On May 06, 2001 at 11:17:37, Heiner Marxen wrote: >On May 06, 2001 at 06:39:02, leonid wrote: > >>On May 05, 2001 at 20:12:25, Heiner Marxen wrote: >> >>>On May 05, 2001 at 18:41:22, Paul wrote: >>> >>>>On May 05, 2001 at 17:47:38, Heiner Marxen wrote: >>>> > >>>>>>>[D]3k4/b2r4/4q3/2q3n1/1nnNbqnN/1Q1qpNBQ/r1BPQPp1/qNQRKR1N b - - >>>>>> >>>>>>And this one took a while longer as usual with your creations :), mate in 8: >>>>>> >>>>>>01:22 BM8 07 exf2+ Nxf2 Qaxd4 Bxd3 Qxf2+ Bxf2 Qxf2+ Rxf2 Bxf2+ Qxf2 Bxd3+ Qe3 >>>>>>g1=Q+ Nxg1 Qf2# >>>>>> >>>>>>Greetings, >>>>>>Paul > >>>Hey, wait a second... while I write this answer, Chest comes up with a mate >>>in 7! Sorry Paul, that is unexpected for me, too. Here we go: >>>There are two key moves: exf2+ and Qaxd4 (the first two moves of your PV). >>>My PVs (with some more variants) look like: >> >> >>Thanks, Heiner! Very appreciated. I will try to solve this position later in 7. >>Was not even sure if mine will have enough time to do this. Had very bad >>branching factor for this one. Just to give you an idea: >> >>3 moves - 0.1 sec >> branching factor - 66 >>4 moves - 6.59 sec >> - 38.6 >>5 moves - 4 min 14 sec >> >>And this is where and why I stopped searching by brute froce the last time. >> >>Cheers, >>Leonid. > >I don't know, why your effective branching factor is so much worse than >mine, but here is my data for comparison: > >depth seconds speed ># 3 0.05 0.98 110- 0 ># 4 1.03 1.05 2162- 0 ># 5 21.94 1.21 52287- 0 ># 6 470.91 1.51 1467975- 11 ># 7 8123.31 1.68 27139229- 18391335 > >As you can see from the "speed" (estimated hash table speed up) the hash >table is not the main difference this time. More data for you to compare: >the executed moves per depth to go: > > black white >mvx 7: 108 133 [108.000 1.231] mvskip lvskip >mvx 6: 2191 2615 [ 16.474 1.194] >mvx 5: 61015 60627 [ 23.333 0.994] 1101 >mvx 4: 1629538 1488284 [ 26.878 0.913] 37238 10 >mvx 3: 30390508 27233557 [ 20.420 0.896] >mvx 2: 409525400 71961657 [ 15.038 0.176] >mvx 1: 84923563 0 [ 1.180 ] > >The defender (white) manages to nearly always check the defender. >Whenever I start to generate legal moves I look, how often the side to >move is in check: > >mg 352166449 W; inchk: 240798413 none 109366513 one 2001523 double >mg 27752191 B; inchk: 4158972 none 23475046 one 118173 double > >Of 27.7 million times generated black moves, black is only 4.1 million times >not in check. > >I know, that you try check moves first. When there are multiple check moves, >do you sort them? There are good checks and bad checks: the goal is to >have more check moves the next time, also. Only in the plys that are "very responsive" previous "best move" is put at the head of the line, if he still existe. If previous "best move" is checking move, it is put at the head of only checking moves. To see what kind of usage of previosly found "best move" is the most sutable can be found only by long statistics. Usage of previous "best one" I started only with my last version. It permitted to speed my brute force and selective search but not all the time. Speeding is comfortable and reasonable but only "statistically speaking". In some position new version even loose in time. My biggest problem with the mate solver is that I never came back (with one exception) to rewrite it. Only some real work make a difference. And when I wrote it, it look just too far ahead of every professional program mate solving capacity to even think about improving it. Just one direct example to give you real idea of what I am been talking about. I will not speak about my selective aearch (it look as good as difficult to compare) but about brute force search that is more comparable. I have no data about this position but one before it (easy 9 moves position) brute force search was: LLchess "One professional program" 4 moves - 0.55 sec 6 sec 5 moves - 0,76 sec 2 min 30 sec 6 moves - 12 sec 53 min 37 sec. Numbers here are typical. At least, now I can see some serious mate solving program from your side to even feel some desire for moving ahead. I hope it will help. Weak mate solving capability in professional programs rely (I guess so) on the obvious fact that mate solver is not in demand. There were so many chess championships in the recent years but probably never for mate solvers. If one day it will be changed, abundance of good mate solvers could be first result of this change. Salut from Montreal! Leonid. >Just an idea... > >Cheers, >Heiner
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.