.
Hi Blue,
Are you sure you posted this in the right thread?
What do you mean by "hidden+code" ?
.
What do you mean by "hidden+code" ?
[hidden][code]
puzzles ...
[/code][/hidden]
blue wrote:More puzzles in "T4".
blue wrote:What do you mean by "hidden+code" ?
- Code: Select all
[code][hidden]
puzzles ...
[/hidden][/code]
I wanted them in hidden "code" text, but it wouldn't work.
denis_berthier wrote:This confirms my previous conclusion that template-depth is not of much use to rate puzzles.
...4.6.....7..92...8..7..152...9.....7..4.....3......8..1.6.59........3......5..2 3.0/3.0/3.0
.2..5..8.4....9...8..3..1....9....6.3....49..7......1.....7.6.8.7......1...2.5.7. 8.8/8.8/8.7
blue wrote:New rules needed ...
1st concept: A (potentially) valid template[k] for digits {d1,...,dk}, should cover every cell with pencilmarks restricted to {d1,...,dk}.
2nd concept: A cell, X, that is covered by every (potentially valid) template[k] for digits {d1,...,dk}, should not have candidates for any other digit.
blue wrote:The bad (?) news: I used a neighborhood search to produce another ~3700 T4 puzles.
Including the ones already posted, they all dropped to T3 or T2, with a rule for the 1st concept.
denis_berthier wrote:I can't remember why I got interested in templates. But I know perfectly well why I lost any interest: because of the various results I reported in previous posts.
denis_berthier wrote:The question is, would the new conditions make template'-depth better correlated to usual measures of complexity (B, BxB, SER...)?
denis_berthier wrote:Did you try what it gives for cbg-000?
cbg-000
-- original
T0 : 7489 (35.04%)
T1 : 3630 (16.98%)
T2 : 1916 ( 8.96%)
T3 : 8340 (39.02%)
-- original + 1st concept
T0 : 7489 (35.04%)
T1 : 3630 (16.98%)
T2 : 5554 (25.98%)
T3 : 4702 (22.00%)
-- original + 1st & 2nd concepts
T0 : 7489 (35.04%)
T1 : 3630 (16.98%)
T2 : 6298 (29.46%)
T3 : 3958 (18.52%)
denis_berthier wrote:(even for size 1, the Tk-assert rule can be skipped)
denis_berthier wrote:Can one find different formulations (apart from the two weaker ones already mentioned in this thread)?
blue wrote:denis_berthier wrote:The question is, would the new conditions make template'-depth better correlated to usual measures of complexity (B, BxB, SER...)?
I'm curious about that myself.
I thought about mentioning the possibility, in the post above.
SER B gB W gW
original 0.5475 0.7160 0.7174 0.7114 0.7139
original+ 0.5839 0.7781 0.7792 0.7743 0.7766
original++ 0.5831 0.7846 0.7857 0.7812 0.7834
blue wrote:These are the "non0-correlation" coefficients (mentioned here) for SER,B,W,gB and gW ratings for the cbg-000 collection, for the 3 rule systems.
(I used weighting appropriate for a "controlled bias" collection).
- Code: Select all
SER B gB W gW
original 0.5475 0.7160 0.7174 0.7114 0.7139
original+ 0.5839 0.7781 0.7792 0.7743 0.7766
original++ 0.5831 0.7846 0.7857 0.7812 0.7834
blue wrote:denis_berthier wrote:Can one find different formulations (apart from the two weaker ones already mentioned in this thread)?
I started writing something, but I realize that it probably wouldn't mean much to your implementation using CLIPS
blue wrote:denis_berthier wrote:(even for size 1, the Tk-assert rule can be skipped)
I noticed that ... that removing it, doesn't change the ratings :
denis_berthier wrote:blue wrote:denis_berthier wrote:(even for size 1, the Tk-assert rule can be skipped)
I noticed that ... that removing it, doesn't change the ratings :
(...)
Define (extended) resolution theory T1 as in the first post of this thread.
Define T1bis as T1 without the T1-assert rule and T2bis = T1bis+T2-delete = T2 without the T1-assert rule.
First, note that there's no way to prove T1-assert within T1bis.
denis_berthier wrote:Note 1: if you wanted to apply a similar line of reasoning to Blue's "concept1" and "concept2", you would need to add at least Subsets[k] to Tk.
blue wrote:denis_berthier wrote:First, note that there's no way to prove T1-assert within T1bis.
It can be "proved".If T1-assert would assert "1r1c1", then
- every template[1] for '1', includes r1c1
- no template[1] for '1', includes a cell that can see r1c1
- T1-delete rules (for digit '1') will eventually produce a hidden single: 1r1c1