The New Sudoku Players' Forum

[champagne]

In the grey area, SER ratings go form 7.5 to 10.8

I defined grey zone with SER between 9.0 and 10.5. These were fuzzy values, but extending them overly makes it meaningless.

It's just a question to find somebody to do the work.

I took a cut off not too expensive as first sample

[champagne]

Hi Champagne

Sampling your puzzle collections I'm having problems with many of them recognising the JExocets your program has found. I suspect this is because it's reporting those where a given in the JE band is included in the base set. Is that right?

If so, how easy would it be for you to either
a) to screen these out
or
b) to screen out those with no eliminations in the target cells?

The JExocet signature pattern isn't that uncommon and will regularly appear in reduced cells in the easier puzzles. However I believe that in these circumstances only rarely will they be of any use.

It would take me ages to do this manually

David

Hi David,

I checked quickly my code and I see no exclusion (in the JE process) for a given in the band.
Your b) is not as such a problem, just a matter of coding and could clear many degenerated forms of JEs.

The missing issue is what I noticed earlier. Working after simple eliminations, the solver sees some JEs not there at the start.

Just to be as fruitful as possible,
1) what is the file you tried to explore? I'll use it to make tests
2) Should I do it before eliminations (with the lay out of my code, I have difficulties to work on the original PM. the minimum would be the very first assignments.)

[eleven]

Now i stopped the generation of pseudo random puzzles with 86 mio puzzles.
I uploaded all puzzles with skfr >=8.6 here (7MB) and about 0.5 mio with skfr between 7.5 and 8.5 here (about 15MB).

It would be nice to see stats for JE's (with eliminations in the target cells) for them.

This is the ratings distribution > skfr 6.1 in absolute numbers found in 86 mio puzzles.

Hidden Text: Show

6.2 26628
6.3 0
6.4 0
6.5 38670
6.6 5739803
6.7 728415
6.8 401201
6.9 390215
7.0 477139
7.1 7265721
7.2 6465558
7.3 1761849
7.4 86534
7.5 55522
7.6 241016
7.7 132269
7.8 116656
7.9 11853
8.0 1435
8.1 46
8.2 167795
8.3 1283097
8.4 683738
8.5 216092
8.6 28374
8.7 21913
8.8 86807
8.9 101911
9.0 41203
9.1 2219
9.2 158
9.3 3

Remark : All puzzles are in T&E(1), which is no surprise. I guess T&E(2) should be rarer than say 9.5 puzzles.

You can see that e.g. more than 90% of all ER9+ puzzles have ER=9.0. So the random JE commonness of all 9+ puzzles (Denis' grey zone) will be very near to that of the ER=9.0 puzzles.
In champagne's list this distribution will be very different. So it would be interesting to compare the JE rates for 2 fixed ratings in both lists (e.g. 8.3 and 9.0), to get a feeling about the bias.

[David P Bird]

Hi Champagne

I have taken very small samples from
List 1 136 Puzzles " here a first list of puzzles with possible things of interest for you"
List 4 489 puzzles " and a lot of puzzles in the grey area with a double exocet 3 digits as in the green area, we have many more double exocets with 4 digits"
I didn't find a single hit in list 1 and had mixed success in list 4.

Here's my thinking:
Low grade JExocets don't provide any immediate eliminations or useful inferences
Medium grade ones don't provide any immediate eliminations but could provide useful inferences
High grade ones provide both immediate eliminations and useful inferences

As I've tried to explain before, if the base set contains digits that are givens in the same band, the information the pattern will give is most often available using simple methods.

The easier the puzzle, the more low-grade JExocets you will find, so don't bother scoring a hit unless there are immediate deletions. However, as the puzzles get harder, even if there are no immediate eliminations, there may still be useful inferences that could be of help, which would be missed by following that policy.

Together these points suggest
1) basic eliminations should be made before the search.
2) For each band determine which digits don't appear as givens and use that set to test for the base and target cells.
3) For green area puzzles only score hits with eliminations in the target cells.
4) For grey area puzzles score all hits.

As you can only do the screening in 3) above, I suggest you process the green area puzzles first.
In the longer term, I think using 2) above could speed up your code tremendously.

Eventually we could then evaluate where the green/grey boundary should be by seeing where the extra inferences help solve puzzles.

My wish list is to split out:
Single JE3s
Single JE4s
Double JE4s
JE+s or twin JEs as you originally called them, where the two object cells contain a Almost Hidden Pair with single locked digit and any combination of base and non-base digits.

I thought you were working towards that, but from your file headings you have mixed them up. However if you add the details of the pattern found as you've done on the ones you posted in the project files, it won’t matter so much as that file can be searched.

I think found one of your Double J3s in your lists today which is interesting as one target cell is common both Exocets.
98.7..6..5...4......3..9...4......5..6.2..7....9..3....1.....67...9...8.....281.. base set (345) tier 3.
However I can see other possibilities too which need to be sorted out!

[champagne]

Now i stopped the generation of pseudo random puzzles with 86 mio puzzles.

Hi eleven

Just 2 questions on that sample

1) are they minimal and ED puzzles
2) do you have the clues distribution

it can be faster than checking on my side

[eleven]

Minimal, yes.
I did not check for equivalences, but it is very unprobable, that there are any in random puzzles.
No, i did not count the clues, but this should not take much time.

[Leren]

David P Bird Wrote: I think found one of your Double J3s in your lists today which is interesting as one target cell is common both Exocets.
98.7..6..5...4......3..9...4......5..6.2..7....9..3....1.....67...9...8.....281.. base set (345) tier 3.
However I can see other possibilities too which need to be sorted out!

I've tested quite a few of Champagne's Grey area double JE3 list and I've noticed that for double JE3's one target cell is always common to both Jexocets.
Nevertheless the double Exocet eliminations work much the same as they do for double JE4's.

eg 98.7..6..7...5.....4.8..7..8....3....279..4....9....1...86...94....278..........6 has Exocets r7c5 r7c6 r9c2 r9c7 135 + r8c8 r8c9 r9c2 r9c4 135 and has 28 double JE eliminations.

Many of the puzzles in this list are so trivial that they solve with basics + a Skyscraper, so whether double JE3's are a legitimate first strike solving technique or just an academic oddity
remains to be determined. Also, I ran across the following position in this puzzle from Champagne's huge JExocets Grey Area database

98.7.......7.65.........7..4...3..2..1......9..95..8..1......4...59..6.......2..3; ;2;0;match type;50;r1c5 r1c6 r3c3 r2c7 1234;r2c1 r2c2 r3c4 r1c7 234;;;4;;

Code: Select all

*--------------------------------------------------------------------------------*
| 9       8       12346    | 7       124     134      |T34      3-5-16  24-5-16  |
|B23     B234     7        | 8       6       5        | 1234    9       124      |
| 2356    23456   1234     |T234     9       34       | 7       1368    12468    |
|--------------------------+--------------------------+--------------------------|
| 4       567     8        | 16      3       9        | 15      2       1567     |
| 23567   1       236      | 246     2478    4678     | 345     3567    9        |
| 2367    2367    9        | 5       1247    1467     | 8       1367    1467     |
|--------------------------+--------------------------+--------------------------|
| 1       23679   236      | 36      578     3678     | 259     4       2578     |
| 2378    2347    5        | 9       1478    13478    | 6       178     1278     |
| 678     4679    46       | 146     14578   2        | 159     1578    3        |
*--------------------------------------------------------------------------------*

This is another case of the move I alluded to on page 11 of this thread

As there are 2 5's in Row 1 (or Box 3) one of r1c89 must be 5 and the other must solve to the same base digit as r4c3 => -16 r1c89
A corollary is that any base digit in r4c3 that isn't in either of r1c89 could be eliminated (didn't occur here but it's possible).

Leren

[denis_berthier]

All puzzles are in T&E(1), which is no surprise. I guess T&E(2) should be rarer than say 9.5 puzzles.

Considering the way SER is defined, it may be difficult to assign a precise boundary. But it was one of the reasons for my broad definition of the grey zone: it includes the T&E(1) vs T&E(2) and the gT&E(1) vs gT&E(2) boundaries. Extending it in any direction beyond the [9.0, 10.5] interval would empty this notion of any meaning.

You can see that e.g. more than 90% of all ER9+ puzzles have ER=9.0. So the random JE commonness of all 9+ puzzles (Denis' grey zone) will be very near to that of the ER=9.0 puzzles.

Before launching long computations for JE on the whole list, I recommend reading my recent post in the old "real distribution of minimal puzzles" thread.

I have computations running for puzzles in the controlled-bias collection, for W ratings >= 9 * . As of now, there's no JE.

(*) this doesn't include all the SER >= 9.0, but I think it's closer to my mental view of the grey zone.

In champagne's list this distribution will be very different. So it would be interesting to compare the JE rates for 2 fixed ratings in both lists (e.g. 8.3 and 9.0), to get a feeling about the bias.

I think the strong bias in champagne's list can easily be seen by examining the distribution of the number of clues.

These results confirm what I thought from the start: all the JEs found by champagne come from the original "potentially hardest" collection. There's nevertheless an interesting new result in his lists (provided that checks are made on degenerated cases): JExocets are not restricted to the topmost hardest.

[champagne]

a quick first comment on eleven work

First of all, I found only 75 puzzles having the exocet pattern in the rand 8.6 file

eleven's sample is only 1/1000 in size of my sample, but I tried to compare the clues pattern

Code: Select all

   "grey area"    pot hard          rand 8.6    
20      889  0.00%     75  0.01%  11.9     2  0.00%
21   636355  0.30%   9789  1.10%  65.0   283  0.10%
22  7925965  3.78%  67540  7.60% 117.4  5814  2.06%
23 39503109 18.84% 232195 26.13% 170.1 36563 12.94%
24 84915032 40.49% 348379 39.20% 243.7 87547 30.98%
25 76709585 36.58% 230588 25.95% 332.7 91557 32.40%
26     3829            10        382.9 46586 16.49%
27     3642            33        110.4 12283  4.35%
28     1485            24         61.9  1794  0.63%
.....
209 699 959       888 658            282 588   

In that table,
the first group is the sample coming out of the potential hardest file
the second group is the original potential hardest file
the third group if eleven's file.

If we accept that so far explored areas in the potential hardest are
deeply 20_23 clues
seriously 24 clues
ongoing 25 clues

and knowing that the sample comes from a +-3 generation, the comparison sample to potential hardest is quite normal.

I put in the fifth column the ratio sample / potential hardest tending to show that the yield is much higher in the area 23 24 clues


Comparing the distribution with eleven's file, we see the effect of the deep search in the 20_22 area, but eleven's file has also the highest density in the area 23 26 clues.

This let think that 26 clues would be the limit for a productive search in potential hardest.

And we are now with the main fact for the JE topic, the ratio in eleven's file is very very low. The most probable reason is that exocets will be found mainly in the vicinity of existing ones.

[champagne]

for those who are interested, the distribution of JEs per number of clue in the last update of my data base (as far as JE tag is concerned)
The count will grow after the last generations mainly for 24 and 25 clues

clues puzzles with JE
20 42
21 1710
22 37204
23 173085
24 229733
25 76665

[denis_berthier]

for those who are interested, the distribution of JEs per number of clue in the last update of my data base (as far as JE tag is concerned)
The count will grow after the last generations mainly for 24 and 25 clues
clues puzzles with JE
20 42
21 1710
22 37204
23 173085
24 229733
25 76665

Out of a total of how many puzzles analysed?

[champagne]

....
Out of a total of how many puzzles analysed?

I am speaking here of the data base of potential hardest (about 80% of JEs) but I 'll give to-day the update of the whole current base (888658 puzzles 76.12% JEs).
.

[denis_berthier]

....
Out of a total of how many puzzles analysed?

I am speaking here of the data base of potential hardest (about 80% of JEs) but I 'll give to-day the update of the whole current base (888658 puzzles 76.12% JEs).

To be clear, I'm not interested in how many puzzles there are in your hardest list, but how many derived ones you checked for JEs.

[champagne]

Here is the last status of my data base of potential hardest.

Code: Select all

20   42
21   1734
22   39724
23   194879
24   281375
25   158948

I have slightly more puzzles than in the last count (676702 against 676487) which reflects either old bugs or a small change in the code.
Usually, I don't revise existing flags unless I have explained why.
I need time to find why

the current base has 888 658, same as for the sample.

[champagne]

98.7.......7.65.........7..4...3..2..1......9..95..8..1......4...59..6.......2..3; ;2;0;match type;50;r1c5 r1c6 r3c3 r2c7 1234;r2c1 r2c2 r3c4 r1c7 234;;;4;;

Code: Select all

*--------------------------------------------------------------------------------*
| 9       8       12346    | 7       124     134      |T34      3-5-16  24-5-16  |
|B23     B234     7        | 8       6       5        | 1234    9       124      |
| 2356    23456   1234     |T234     9       34       | 7       1368    12468    |
|--------------------------+--------------------------+--------------------------|
| 4       567     8        | 16      3       9        | 15      2       1567     |
| 23567   1       236      | 246     2478    4678     | 345     3567    9        |
| 2367    2367    9        | 5       1247    1467     | 8       1367    1467     |
|--------------------------+--------------------------+--------------------------|
| 1       23679   236      | 36      578     3678     | 259     4       2578     |
| 2378    2347    5        | 9       1478    13478    | 6       178     1278     |
| 678     4679    46       | 146     14578   2        | 159     1578    3        |
*--------------------------------------------------------------------------------*

This is another case of the move I alluded to on page 11 of this thread

Leren

I did not follow your point page 11, but that puzzle, even if not seen as such by my solver is a double exocet.

r1c3=6
r3c89 <234>