Author: Heiner Marxen
Date: 14:54:10 01/19/01
Go up one level in this thread
On January 19, 2001 at 17:06:44, leonid wrote: >On January 19, 2001 at 15:32:49, Heiner Marxen wrote: > >>On January 18, 2001 at 22:06:37, leonid wrote: >> >>>On January 18, 2001 at 19:13:22, Paul wrote: >>> >>>>On January 18, 2001 at 19:01:08, Pete Galati wrote: >>>> >>>>>On January 18, 2001 at 18:44:05, leonid wrote: >>>>> >>>>>>Hi! >>>>>> >>>>>>If you would like to find a mate here is one position. It is very easy to solve >>>>>>but not that simple to find shortest mate. >>>>>> >>>>>>I failed here even in 3 hours to find shortest mate through brute force search. >>>>>>Maybe you will have better chance. >>>>>> >>>>>> 1k4q1/1pppPr2/PbbP1N1n/QP2rn1R/1q6/1q2RBB1/q1q2PPP/6NK black to go >>>>>> >>>>>>Leonid. >>>>> >>>>>Is this the right position? >>>>> >>>>>[D]1k4q1/1pppPr2/PbbP1N1n/QP2rn1R/1q6/1q2RBB1/q1q2PPP/6NK b - - 0 1 >>>> >>>>If so, then mine says it's a mate in 10 for black starting with Nxg3+ ... >>>>It looks like a daily reverse auction ... Anyone lower, anyone? :) >>>> >>>>Paul >>> >>>Position is exact and response 10 moves must be or shortest possible, or very >>>close move. Actually my program solved this position through selective search in >>>11 moves. Solution was instant. Finding shortest mate is the other story. >>>Through brute force program said in 8 mate is not there. Only 9 still could be >>>looked. >>> >>>I found that branching factor was terrible to see 9 moves. AMD 400Mhz. >>> >>>8 moves - 2h 44 min. >>>7 moves - 2 min 48 sec. >>>6 moves - 7 sec. >>> >>>Leonid. >> >>Now, with the correct side to move, Chest has still found "no mate in 9". >>But the timing indicate an unusually large branching factor: >># 1 0.00 0.87 1- 0 >># 2 0.01 1.00 1- 0 >># 3 0.02 0.96 89- 0 >># 4 0.06 1.09 437- 0 >># 5 0.27 1.30 1693- 0 >># 6 2.04 1.38 6662- 0 >># 7 31.09 1.49 63238- 0 >># 8 824.56 1.75 1256916- 3 >># 9 24061.56 2.19 37677207- 28929306 >>(depth, seconds, speed, nodes in-out) on a K7/600 (350 MB TT) >> >>That are already 6.7 hours. I'm not sure I will wait until the mate in 10 >>arrives. We can expect over 8 days from the above data. The last two >>lines have indicate an effective branching factor of 29.2. >>Although this is a bit better than Leonid's factor 60, it is still quite >>a bit too heavy. >> >>Heiner > >So it was really 10 moves positions! > >Your branching factor is much better that mine. Only position? Better move >ordering? I don't know. It could be even hash table. And how after your >experience hash helps in branching factor? If there ever existe some difference >at all. > >Can see that in general our branching factor have the same tendency for this >position. It grows to the worst with the number of plys to be seen. In mine it >goes this way: > >4 moves - 0.1 sec > branching = 8.8 0.27 / 0.06 = 4.5 >5 moves - 0.88 sec > branching = 8.2 2.04 / 0.27 = 7.55 >6 moves - 7.25 > branching = 23 31.09 / 2.04 = 15.24 >7 moves - 2 min 47 sec > branching = 59 824.56 / 31.09 = 26.52 >8 moves - 2 h 43 min 56 sec > >Leonid. Since increasing depth seems to help me as compared to you, I suspect much of it is caused by the TT. My hit rate is not large, 8.6% normal hits and 18.2% hits for ETC (enhanced transposition cutoffs). But with some depth left to go that can have quite an effect. To measure the effect of good move ordering, I would have to replace it with another one, and re-run. I'm too lazy for this just now. I suspect that you will at least also prefer check moves and captures, so for this special problem we may do comparatively equal: white has left quite some pieces to be agressive in that simple way. BTW, here are my branching factors inside the "no mate in 8": mvx 8: 87 87 [ 87.000 1.000] 1 mvx 7: 438 533 [ 5.034 1.217] mvx 6: 1713 2259 [ 3.214 1.319] 1 mvx 5: 7444 9227 [ 3.295 1.240] 154 4 mvx 4: 70334 74945 [ 7.623 1.066] 2052 5 mvx 3: 1469915 1314821 [ 19.613 0.894] mvx 2: 33639291 23346271 [ 25.585 0.694] mvx 1: 4117503 0 [ 0.176 ] Just implement your TT. You will love the effect! Heiner
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.