denis_berthier wrote:Red Ed wrote:What are "p. 155" and "p. 160"? A reference to a book?
My last book
Unfortunately i could not afford it - there were so much other books, i had to buy.Red Ed wrote:That's one impressive tome!
denis_berthier wrote:Red Ed wrote:What are "p. 155" and "p. 160"? A reference to a book?
My last book
Unfortunately i could not afford it - there were so much other books, i had to buy.Red Ed wrote:That's one impressive tome!
eleven wrote:Unfortunately i could not afford it - there were so much other books, i had to buy.
Number of solution grids: 21457975
Number of 30-clue subgrids that were proper: 215645
Total number of minimal puzzles in those 215645 30-clue subgrids:
+----+----------+--------------+------------+
| Cl | Count | E(nr/grid) | E(rel err) |
+----+----------+--------------+------------+
| 23 | 2 | 8.8918e+012 | 70.71% |
| 24 | 8 | 8.5954e+013 | 35.36% |
| 25 | 29 | 7.1041e+014 | 18.57% |
| 26 | 21 | 1.1080e+015 | 21.82% |
| 27 | 14 | 1.5047e+015 | 26.73% |
| 28 | 7 | 1.4510e+015 | 37.80% |
+----+----------+--------------+------------+
Number of solution grids: 694676
Number of 30-clue subgrids that were proper: 7154
Total number of minimal puzzles in those 7154 30-clue subgrids:
+----+----------+--------------+------------+
| Cl | Count | E(nr/grid) | E(rel err) |
+----+----------+--------------+------------+ Linear probing
| 20 | 20 | 4.4984e+006 | 36.06% | E(rel err) for
| 21 | 1124 | 1.5421e+009 | 11.55% | comparison ...
| 22 | 19257 | 1.7614e+011 | 5.42% |
| 23 | 106227 | 7.1658e+012 | 2.80% | 70.71%
| 24 | 190937 | 1.0672e+014 | 1.94% | 35.36%
| 25 | 119587 | 6.3499e+014 | 1.73% | 18.57%
| 26 | 25332 | 1.5065e+015 | 2.15% | 21.82%
| 27 | 1797 | 1.4694e+015 | 4.70% | 26.73%
| 28 | 41 | 6.0348e+014 | 19.66% | 37.80%
| 29 | 2 | 7.8010e+014 | 70.71% |
+----+----------+--------------+------------+
Red Ed - Posted: Wed Oct 14, 2009 6:59 am p. 37 of the pdf files - wrote:On the parts of the distribution that we're each focussing on, it seems we each have a method tuned for the purpose: my supersets method has no hope of beating yours below 30 clues (nor does it try to); and your paths method has no hope of beating mine above 30 clues (nor do you proclaim any interest in doing).
Really? In what sense does anything I've just said relate to the supersets method?denis_berthier wrote:How miraculous that, after only 10 minutes of computation with the same program, everything is changed!Red Ed - Posted: Wed Oct 14, 2009 6:59 am p. 37 of the pdf files - wrote:On the parts of the distribution that we're each focussing on, it seems we each have a method tuned for the purpose: my supersets method has no hope of beating yours below 30 clues (nor does it try to); and your paths method has no hope of beating mine above 30 clues (nor do you proclaim any interest in doing).
Sadly, this doesn't help, due to the overhead of reaching the prune-points in the first place. Oh well.Red Ed wrote:It wouldn't be hard to adapt the depth-first searcher to run a combined strategy (you just prune the search space appropriately), giving the best of all worlds.
Red Ed wrote:Really? In what sense does anything I've just said relate to the supersets method?denis_berthier wrote:How miraculous that, after only 10 minutes of computation with the same program, everything is changed!Red Ed - Posted: Wed Oct 14, 2009 6:59 am p. 37 of the pdf files - wrote:On the parts of the distribution that we're each focussing on, it seems we each have a method tuned for the purpose: my supersets method has no hope of beating yours below 30 clues (nor does it try to); and your paths method has no hope of beating mine above 30 clues (nor do you proclaim any interest in doing).
denis_berthier wrote:If it doesn't, what does it relate to in this thread?
Red Ed wrote:Let that be an exercise for the readerdenis_berthier wrote:If it doesn't, what does it relate to in this thread?
Red Ed wrote:Please, Denis, just read the opening post.
Red Ed - Posted: Wed Oct 14, 2009 6:59 am p. 37 of the pdf files - wrote:On the parts of the distribution that we're each focussing on, it seems we each have a method tuned for the purpose: my supersets method has no hope of beating yours below 30 clues (nor does it try to); and your paths method has no hope of beating mine above 30 clues (nor do you proclaim any interest in doing).
Nope, I've not revisited supersets since 2009, so I suppose the conclusions about >30 clue minimals are unchanged; but I don't understand why you think that part of the problem space is relevant to the results I just reported.denis_berthier wrote:(quoting observations on the supersets method ...) Do you have anything new that could change this old conclusion ?
Red Ed wrote:I hope Blue won't mind me opining that the major contribution might be more the method than the commentary.
Red Ed wrote:denis_berthier wrote:Red Ed - Posted: Wed Oct 14, 2009 6:59 am p. 37 of the pdf files - wrote:On the parts of the distribution that we're each focussing on, it seems we each have a method tuned for the purpose: my supersets method has no hope of beating yours below 30 clues (nor does it try to); and your paths method has no hope of beating mine above 30 clues (nor do you proclaim any interest in doing).
Do you have anything new that could change this old conclusion ?
Nope, I've not revisited supersets since 2009
denis_berthier wrote:OK, let me know if some day you have anything new.
in his Wed Jun 19, 2013 3:55 pm edit, denis_berthier wrote:Red Ed wrote:These methods were described originally back in 2009, in the context of counting minimal puzzles
This is an obvious falsification of history. If the "subset method" had been described four years ago, how could anyone believe you didn't run it 10 minutes
Red Ed wrote:in his Wed Jun 19, 2013 3:55 pm edit, denis_berthier wrote:Red Ed wrote:These methods were described originally back in 2009, in the context of counting minimal puzzles
This is an obvious falsification of history. If the "subset method" had been described four years ago, how could anyone believe you didn't run it 10 minutes, instead of speaking only of the "superset method" and admitting thatRed Ed - Posted: Wed Oct 14, 2009 6:59 am p. 37 of the pdf files - wrote:On the parts of the distribution that we're each focussing on, it seems we each have a method tuned for the purpose: my supersets method has no hope of beating yours below 30 clues (nor does it try to); and your paths method has no hope of beating mine above 30 clues (nor do you proclaim any interest in doing).
Hello again. May I refer you to my post of Sat Jul 18, 2009 2:15 pm on RDMP page 11?
Denis, I am running code from 2009. It starts:denis_berthier wrote:OK, you had some definition, but it doesn't explain why you didn't run it and didn't even mention it in your overall conclusion recalled above. Part of your current algorithm must be recent.
/* minsubs.c : simple DFS subsets method for the minimals-counting problem
*
* This version placed on www.sudoku.com on 23 December 2009
The "generic method" is just a convenient framework within which to explain the practical applications (subsets method, etc) and, in and of itself, needs no further definition of the distribution underpinning A.BTW, your "generic method" remains devoid of any content without a probability distribution being first defined on P(M). This is so basic probability theory that I can't understand you persist refusing to see it.