Author: leonid
Date: 11:07:00 07/18/01
Go up one level in this thread
On July 18, 2001 at 06:44:37, Heiner Marxen wrote:
>On July 17, 2001 at 22:09:08, leonid wrote:
>
>>On July 17, 2001 at 19:27:49, Heiner Marxen wrote:
>>
>>>On July 17, 2001 at 08:33:49, leonid wrote:
>>>
>>>>Hello!
>>>>
>>>>This position you can try with every program. Its number of moves is only 89.
>>>>
>>>>[D]k1qnr3/1qq5/qn2Q3/qN1QqQ1K/qN2QqQ1/RbQQqQ2/1RrbQ3/2BB4 w - -
>>>>
>>>>Please indicate your result.
>>>>
>>>>Thanks,
>>>>Leonid.
>>>
>>>Hi Leonid!
>>
>>Hi, Heiner!
>>
>>>This one is not as easy to solve for Chest, as usual. After 2.6 hours on a
>>>K7/600 with 350 MB hash it just found "no mate in 11". The effective
>>>branching factor has climbed from below 4 to above 10, so the next depth
>>>most probably will need more than a day :-( Hence I stop here.
>>
>>You are already there. Since you found mate in 11, it is mate in 12. My
>>selective found mate in 12.
>
>Fine! From the increasing EBF I suspected already to be near the mate
>(increased EBF does often occur in last depth with mate, and sometimes one
>depth before it).
>
>>Was able to reach only 10 moves deep by brute force. It took already 11 hours
>>and 17 min. I must for sure one day install my hash and see the difference. I
>>hope in few months from now to have my Linux computer and start writing once
>>again. For now I do almost nothing useful.
>
>See estimated factor reached by hash in Chest below as "speed".
>
>>My branching factor, as it happened very often, have the same tendency as your.
>>It was 5, between 4 and 5 moves and ended by 9.2, between 9 and 10 moves.
>
>The exact data to compare:
>
> depth time EBF[T] EBF[N] speed
># 1 0.00s 0kN 0.87 1- 0
># 2 0.00s 0kN 1.00 1- 0
># 3 0.02s 1kN [ 8.43] 0.94 90- 0
># 4 0.09s [ 4.50] 4kN [ 5.51] 1.06 521- 0
># 5 0.37s [ 4.11] 16kN [ 3.82] 1.36 2000- 0
># 6 1.27s [ 3.43] 55kN [ 3.35] 1.62 6781- 0
># 7 4.84s [ 3.81] 213kN [ 3.90] 2.20 24475- 0
># 8 17.89s [ 3.70] 896kN [ 4.20] 3.05 84227- 0
># 9 111.61s [ 6.24] 6190kN [ 6.91] 3.23 519135- 0
># 10 870.30s [ 7.80] 47882kN [ 7.74] 3.46 4210571- 9399
># 11 9223.60s [ 10.60] 502283kN [ 10.49] 3.40 45603940- 36856039
>
>The effect of the hash table is estimated to speed up by a factor of slightly
>above 3. That is not dramatic, but quite a difference.
This is very useful to know! I looked into Rebel to have some idea about this
position and had impression that advantage grows with the depth. For instance,
on Rebel 10 with and without 20M of hash.
4 moves - 2 min 46 sec. With hash - 1 min 29 sec.
5 moves - Went more that 3 hours and half and disconnected. Had no more
patience. With hash only 54 min.
Hash for 5 moves gave more advantage that for 4 moves.
>
>Since the estimate sometimes is quite inaccurate, I have run it to depth 8
>with hash completely disabled, and found:
>
># 3 0.01s 1kN [ 8.43] 1.00 0- 0
># 4 0.10s [ 10.00] 5kN [ 6.27] 1.00 0- 0
># 5 0.53s [ 5.30] 27kN [ 5.50] 1.00 0- 0
># 6 2.74s [ 5.17] 134kN [ 5.01] 1.00 0- 0
># 7 14.30s [ 5.22] 700kN [ 5.22] 1.00 0- 0
># 8 68.14s [ 4.77] 3405kN [ 4.86] 1.00 0- 0
>
>14.30 / 4.84 = 2.954 > 2.20
>68.14 / 17.89 = 3.808 > 3.05
Here my program had for sure worst branching factor.
4 moves - 0.32 sec
5 - branching factor
5 moves - 1.64 sec
5.4
6 moves - 8.9sec
7.22
7 moves - 64 sec
7.19
8 moves - 7 min 40 sec
9.58
9 moves - 1h 13m 27sec
9.2
10 moves - 11h 17m 8sec
Heiner, and do you have your branching factor for previous position without
hash? If you could, send it to me, or indicate here. It is for sake of seeing if
my branching factor is permenantly worst for brute force, or it is something
that depend on given position. It could be that my code could be improved much
more that I initially expected. For now my guess is between 3 and 5 time
counting 5 moves brute force search.
Previous position was mate in 8 moves. For "maybe", will indicate my time for
brute force.
4 moves - 0.49sec
8.77
5 moves - 4.34 sec
9.38
6 moves - 40.7 sec
9.35
7 moves - 6 min 20 sec
8 moves, first mate found in 87 sec.
Cheers,
Leonid.
>The real speed up appears to be even a bit larger than estimated.
>
>Cheers,
>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.