The New Sudoku Players' Forum

[champagne]

Upia family start

This is the second family of tridagon vicinity expansion.
I had to define how to process before releasing it, and this is I hope the final way to report on this work.

Just to be clear, this is an attempt to see if we can grow from a unique seed and keep big tridagon families with no bridge to other families.

The 2 ways to grow from a seed will be

The “classical” vicinity search as defined for me in this thread Vicinity search for high ratings : Software

A variant of the Maxexpand where clues can be added as long as the “easy filter” is still active.
========================================
The start for a family is done here is in 2 steps

First is a classical vicinity on all the puzzles of the solution grid seed (here 733 puzzles). This step opens the door to other solution grids
Second is a max expand step on the results of the first one (keeping only puzzles rating skfr >= 10.5)


At the end of these 2 steps, we have
A list of puzzles in canonical morph as this one

[list=]
....56......7.....7891.....21..9.87..4......19......42...2...17.7.9..4.88.1....9.;105;30359916;m01FOQ2Cm8Mf52;356;2

....56......7.....7891.....21..9.87..4......19......42...2...17.7.9..4.88.1....9.; puzzle in minlex solution grid morph
105; skfr rating x 10 (here 10.5)
30359916; solution grid rank
m01FOQ2Cm8Mf52; puzzle bitfield in the solution grid
356; digits of the tridagon threat (could be 0 if the quick code did not find the 3 digits
2 “corner box” box not in the bands/stacks of the tridagon threat[/list]

The entire list will be stored on my google drive, one file per family

In this thread, I’ll give only summaries

[list=]Puzzles per rating,
Puzzles per solution grid
Tridagon threat list here 30359916;356;2 (This is the quickest way to see if we have bridges to another family.)
[/list]

In fact, “solution grid+digit+corner box” defines a tridagon threat, valid also if the solution grid has no tridagon pattern.

Next post contains the summary of the start for the “upia” family

[champagne]

The "upia" family starts from the solution grid rank 30384643 with 12 grids rating 11.8 attached and 733 puzzles

Hidden Text: Show

Code: Select all

.2.4.6............789........53..87...8....92......3.53.2.7.95...7.3.2.8.9......7;118;30384643;g007WP46eLZg24
.2.4.6............789........53..87...8....92.7....3.53.2.7.95.....3.2.8.9......7;118;30384643;g007WP4MeL3g24
.2.4.6............789........53..87..38....92........5..2.7.95...7.3.2.8.9.....37;118;30384643;g007WP66WKZg26
.2.4.6............789........53..87..38....92.7......5..2.7.95.....3.2.8.9.....37;118;30384643;g007WP6MWK3g26
.2.4.67...........789........53..8....8....92.7....3.53.2.7.95.....3.2.8.9......7;118;30384643;g107W94MeL3g24
.2.4.67...........789........53..8....8....92.7....3.53.2.7.95...7.3.2.8.9.....3.;118;30384643;g107W94MeLZg22
.2.4.67...........789........53..8...38....92.7......5..2.7.95.....3.2.8.9.....37;118;30384643;g107W96MWK3g26
.2.4.67...........789........53..8...38....92.7......5..2.7.95...7.3.2.8.9.....3.;118;30384643;g107W96MWKZg22
.2.4.67...........789........53..87...8....92......3.53.2.7.95...7.3.2.8.9.......;118;30384643;g107WP46eLZg20
.2.4.67...........789........53..87...8....92.7....3.53.2.7.95.....3.2.8.9.......;118;30384643;g107WP4MeL3g20
.2.4.67...........789........53..87..38....92........5..2.7.95...7.3.2.8.9.....3.;118;30384643;g107WP66WKZg22
.2.4.67...........789........53..87..38....92.7......5..2.7.95.....3.2.8.9.....3.;118;30384643;g107WP6MWK3g22


after the 2 steps, we get 47048 puzzles (rating >=10.5)
with the following summary

Code: Select all

Er   CountOfpuz
105   34412
106   6405
107   388
108   515
109   554
110   1932
111   2471
112   272
113   30
114   12
116   17
117   28
118   12

we see that we have here no new 11.8
results are in 20 solution grids

Code: Select all


rank   CountOfpuz
30359916   3773
30371007   1103
30383362   1692
30384643   16867  upia
31495381   1506
31681493   56
31693641   869
162014823   1404
190273228   840
516294230   5120
598849128   4466
655525596   6581
662545007   626
685824016   277
689526518   30
910366057   576
910366101   780
1028350500   328
2277668623   93
3651788238   61


"upia" solution grid has now 16867 puzzles (733 at the start)
and for the bridge analysis, one threat per solution grid

Code: Select all


rank   digits   boxbase
30359916   356   2
30371007   146   2
30383362   146   2
30384643   146   2
31495381   146   2
31681493   146   2
31693641   146   2
162014823   458   1
190273228   245   2
516294230   235   1
598849128   458   1
655525596   458   1
662545007   467   1
685824016   128   2
689526518   134   3
910366057   236   3
910366101   236   3
1028350500   129   3
2277668623   356   3
3651788238   248   9

The entire file is not yet in the google drive, I intend to upload it after the next cycles

[denis_berthier]

.
I fail to see anything new here and in particular no "tridagon family" that would have anything special wrt to mith's collection.
What I see concretely is a small bunch of 12 puzzles.
None of them is in T&E(3). They're all in T&E(2) and none of them has BxB > 5.
All of them have a tridagon with a single guardian and all of them can be solved with at worst a finned X-wing (+ the tridagon).
.

[champagne]

.
I fail to see anything new here and in particular no "tridagon family" that would have anything special wrt to mith's collection.
What I see concretely is a small bunch of 12 puzzles.
None of them is in T&E(3). They're all in T&E(2) and none of them has BxB > 5.
All of them have a tridagon with a single guardian and all of them can be solved with at worst a finned X-wing (+ the tridagon).
.

Funny comment on this work. Is missing here the reading of the “target of the project”.

no "tridagon family" that would have anything special wrt to mith's collection.

The main property of a new family is to be
1) Hard in serate well known rating,
2) Not linked to mith’ file.

I am sure that mith have seen many of such puzzles not published due to the “T&E3” check.

I have nothing against the T&E3 property, I am just not interested in it as I am not fun of the back door property. They have no link with the solving paths.

I discarded more than 50% of mith’s file for my project. The easiest of them had a rating 4.5, just above a puzzle for beginners.

My cutoff is skfr 10.5, likely too easy for skill players, but close to the cutoff used in the potential hardest file built many years ago.

The 12 puzzles shown here (out of a file of 47000) comes from my direct scan of solution grids to find “non degenerated pattern”, they are supposed to be good seeds for a vicinity search, that’s it.

None of them is in T&E(3). They're all in T&E(2) and none of them has BxB > 5.

I think that I understood what is T&E(3). In fact, In Sudoku Explainer (2005), Nicolas Juillerat had a “nested one” and a “nested 2” levels very similar in the principles in my opinion, but with a wider set of rules.

When I cloned it in SKFR, I did not implement the “double nested level”. All puzzles are solved by SKFR. This means that the last level in Sudoku Explainer has never been reached.

And I cannot comment on

none of them has BxB > 5.

I just don’t know what is BxB

[blue]

Hi champagne,

Here's a short path ... {-1,+2}, {-2,+2}, {-2,+2} ... from your 8th puzzle (of 12), to one of mith's puzzles.

Code: Select all

.2.4.67...........789........53..8...38....92.7......5..2.7.95...7.3.2.8.9.....3.;118;30384643;g107W96MWKZg22

                                                                                            PGE -M         SE
.2.4.67...........789........53..8...38....92.7......5..2.7.95...7.3.2.8.9.....3. T&E(2) 11.7/1.2/1.2 11.7/1.2/1.2
.2.4.67............891.......53..8...38..7.92.7......5..2.7.95...7.3.2.8.9.....3. T&E(2) 10.4/1.2/1.2 10.4/1.2/1.2
.2.4.67......8.....891.......53......38..7.92.7......5..2...95...7.3.2.8.9..5..3. T&E(2) 10.1/1.2/1.2  9.3/1.2/1.2
.2.4.6.......8.3...891.......53......38..7.92.7.........2...95...7.3.2.8.9..5..37 T&E(3) 11.7/1.2/1.2 10.3/1.2/1.2

The intermediate puzzles have a tridagon (after T&E(1)), but they don't pass your ratings cutoff.
Neither does the T&E(3) puzzle, for that matter.

[champagne]

congratulations,

this is likely not done in blind mode,
we bridge from solution grid rank 30384643
to solution grid rank 1104179066
through solution grid rank 95258249

(I did not check yet if this is always in min lexical mode)

I am curious to know how you selected the source and the target.

This shows that bridge could be a common situation, hidden by the filters (T&E3 skfr >=10.5)...

The problem in blind mode is that the vicinity gives millions of puzzles to rate despite all checks done to avoid redundancy.

How you chose the source/target would be interesting.

So far, I have a lot of work to close ongoing tasks to fill all the time that I can allocate to this project.

[champagne]

Hi champagne,

Here's a short path ... {-1,+2}, {-2,+2}, {-2,+2} ... from your 8th puzzle (of 12), to one of mith's puzzles.

Code: Select all

.2.4.67...........789........53..8...38....92.7......5..2.7.95...7.3.2.8.9.....3.;118;30384643;g107W96MWKZg22

                                                                                            PGE -M         SE
.2.4.67...........789........53..8...38....92.7......5..2.7.95...7.3.2.8.9.....3. T&E(2) 11.7/1.2/1.2 11.7/1.2/1.2
.2.4.67............891.......53..8...38..7.92.7......5..2.7.95...7.3.2.8.9.....3. T&E(2) 10.4/1.2/1.2 10.4/1.2/1.2
.2.4.67......8.....891.......53......38..7.92.7......5..2...95...7.3.2.8.9..5..3. T&E(2) 10.1/1.2/1.2  9.3/1.2/1.2
.2.4.6.......8.3...891.......53......38..7.92.7.........2...95...7.3.2.8.9..5..37 T&E(3) 11.7/1.2/1.2 10.3/1.2/1.2

The intermediate puzzles have a tridagon (after T&E(1)), but they don't pass your ratings cutoff.
Neither does the T&E(3) puzzle, for that matter.

Back to this , I tried to figure out how “blue” could find it.

The task seems easy if you just take both ends

Code: Select all

.2.4.67...........789........53..8...38....92.7......5..2.7.95...7.3.2.8.9.....3. T&E(2) 11.7/1.2/1.2 11.7/1.2/1.2
.2.4.6.......8.3...891.......53......38..7.92.7.........2...95...7.3.2.8.9..5..37 T&E(3) 11.7/1.2/1.2 10.3/1.2/1.2


You have a trivial bridge -5+6 a little too big to be considered as a normal vicinity step.
You just look to break it without the usual filter

The hidden problem is that you can have other morphs of the puzzle in the database.

Here, in my base, the canonical status for the four puzzles is

Code: Select all

.2.4.67...........789........53..8...38....92.7......5..2.7.95...7.3.2.8.9.....3.;118;30384643;g107W96MWKZg22;146;2
.2.4.67............891.......53..8...38..7.92.7......5..2.7.95...7.3.2.8.9.....3.;102;30384643;g10EW9cMWKZg22;146;2
...4.6.8.4.....1.3..9...5............94..72.86.892...7.628........27.9....7.64...;93;95258249;eAe410cjZE0Bq0;135;3
...4.6.8.4..18....6...37............7.16.3..483..4.6.73.....5....8...9.2....18.7.;102;1104179066;eA3n00jSg1Xem2;259;9


So “blue” has another secret weapon to solve this morphing step
 

[blue]

Hi champagne,

Unfortunately, I don't have a secret weapon.
I just got lucky with a "scatter gun" approach, and 3 itterations of "vicintity" expansion.
I had a filter: puzzles must have a tridagon pattern after T&E(1) work.

12 puzzles, 1 grid
--> 4323 new puzzles on 199 grids, none in T&E(3)
--> 440179 new puzzles on 21092 grids, none in T&E(3)
--> 6 puzzles in T&E(3), on 4 grids ... one in mith's list.

The last itteration took ~362 core hours, and would have produced millions of T&E(2) puzzles.

[champagne]

Hi « blue »,

Hi champagne,

Unfortunately, I don't have a secret weapon.
I just got lucky with a "scatter gun" approach, and 3 itterations of "vicintity" expansion.
I had a filter: puzzles must have a tridagon pattern after T&E(1) work.

12 puzzles, 1 grid
--> 4323 new puzzles on 199 grids, none in T&E(3)
--> 440179 new puzzles on 21092 grids, none in T&E(3)
--> 6 puzzles in T&E(3), on 4 grids ... one in mith's list.

The last itteration took ~362 core hours, and would have produced millions of T&E(2) puzzles.

Each test delivers interesting points.

Could you post the other T&E3 or 11.x x>=7 puzzles

==============================
This “local” vicinity with a low filter shows a bridge that my process will not produce.
The target is out of my cutoff, but this is true also for the relay puzzles.

The 17 clues search delivered all 17 puzzles through what mladen dobritchev called “indirect search”.
Here, looking at what happens, it could be possible that a small number of seeds without filters as T&E3 and Er10.5 cover all the field.

Unhappily, this is as you say billions of puzzles to process. So to get such puzzles with high rating, the cutoff and the redundancy control must be used.

===========================

By chance, you expanded only one tridagon as seed and did not (I think) change the format of the puzzles found for the next cycles.

This is not realistic on a bigger scale but gave this nice sequence where the path is very easy to follow.
And raised another point:

Having 2 puzzles as your 2 ends in my “canonical format”

Code: Select all

.2.4.67...........789........53..8...38....92.7......5..2.7.95...7.3.2.8.9.....3.;118
...4.6.8.4..18....6...37............7.16.3..483..4.6.73.....5....8...9.2....18.7.;102


Can we define a quick process to show that they are close to each other (here -5+6) in a vicinity process.

If we can do this, we have another way to find bridges.

=================

Prior to doing more cycles on “upia”, what will be my next task to see if I can find the bridge with the filter on high ratings, I have to close the current vicinity run on lower ratings of my direct scan. This is about 100 millions puzzles to process. The vicinity is nearly over, but remains the redundancy control, the rating …

The run on the highest ratings excluding the families gave only one solution grid with the rating 11.8

[blue]

Could you post the other T&E3 or 11.x x>=7 puzzles

The puzzles that I saved (4323+440179) all had SKFR ratings <= 11.5
FWIW: only (875+12326) had ratings >= 10.5

The six T&E(3) puzzles were:

Code: Select all

..34.67......8.2.....72....2.4.67..8.6........7834...2.42...95.836.....4.........
..34.67....7.8.2......2....2.4.67..8.6.........834...2742...95.836.....4.........
..34.67....7.8.2......2....2.4.67..8.6........7834...2.42...95.836.....4.........
...4.6...4.....12..8....5...4.87.39..........89.6.3.7..7.9.....6..3.7...93..48...
1.3.5.7.9.5......669..7.1...619..3.55........73....9................26.1...8.4..3
...4.6.8.4..18....6...37............7.16.3..483..4.6.73.....5....8...9.2....18.7. (mith's puzzle)

===========================

By chance, you expanded only one tridagon as seed and did not (I think) change the format of the puzzles found for the next cycles.

This is not realistic on a bigger scale but gave this nice sequence where the path is very easy to follow.

I had the parh, originally in canonical form puzzles.
I reverse engineered it, by hand, to produce the nicer looking path.

Having 2 puzzles as your 2 ends in my “canonical format”

Code: Select all

.2.4.67...........789........53..8...38....92.7......5..2.7.95...7.3.2.8.9.....3.;118
...4.6.8.4..18....6...37............7.16.3..483..4.6.73.....5....8...9.2....18.7.;102


Can we define a quick process to show that they are close to each other (here -5+6) in a vicinity process.

If we can do this, we have another way to find bridges.

It seems like a long time job, just for two puzzles.
Even if it could be done in a few msec, to check one puzzle against mith's entire file, would take hours.

Years ago, I wrote something to consider all morphs of the 2nd puzzle, to find the best "hamming distance" between two inputs.
I'll see if I can find the code, and look into how fast it was.
It might be possible to modify it to produce a result like {-5,+6}.
Still though, knowing that one puzzle is {-5,+6} away from a particular morph of a 2nd puzzle, wouldn't guarantee a short path using small {-n,+m}'s.

Best regards,
Blue.

[blue]

Years ago, I wrote something to consider all morphs of the 2nd puzzle, to find the best "hamming distance" between two inputs.
I'll see if I can find the code, and look into how fast it was.
It might be possible to modify it to produce a result like {-5,+6}.
Still though, knowing that one puzzle is {-5,+6} away from a particular morph of a 2nd puzzle, wouldn't guarantee a short path using small {-n,+m}'s.

I found the code.
It operated on full grids, instead of puzzles.
I was able to modify to handle puzzles, and produce a {-m,+n} result.
Bad news: it takes ~200ms at 4.3GHz, to compare two puzzles

[champagne]

After your comments, I was not expecting a miracle...

I am just wondering if limiting the search to puzzles having a tridagon threat could make it feaseable.

In a canonical view, the box one can be set to the "corner box" of the pattern.

But as soon as the pattern has many guardians (after easy paths) the morphing remains difficult.

Not very exciting, I'll see what comes out of the vicinity with my 10.5 filter.