The New Sudoku Players' Forum

[yzfwsf]

The score given by SE may be too high.

Code: Select all

 *-----------*
 |..6|4..|5..|
 |...|.23|.9.|
 |8..|..5|..2|
 |---+---+---|
 |7..|..4|2..|
 |..5|3..|.7.|
 |...|.1.|..3|
 |---+---+---|
 |.4.|9..|...|
 |..8|...|...|
 |1..|...|8.6|
 *-----------*
..64..5......23.9.8....5..27....42....53...7.....1...3.4.9.......8......1.....8.6


[Leren]

Code: Select all

*--------------------------------------------------------------*
| 239  12379 6     | 4       789    189-7  | 5     138   178   |
| 45   157   147   | 1678    2      3      | 1467  9     1478  |
| 8    1379  13479 | 167     679    5      | 13467 1346  2     |
|------------------+-----------------------+-------------------|
| 7    13689 139   |c568    c5689   4      | 2     168-5 189-5 |
| 2469 12689 5     | 3      c689   a2689   | 1469  7     1489  |
| 2469 2689  249   | 27-568  1     a26789  | 469   4568  3     |
|------------------+-----------------------+-------------------|
| 2356 4     237   | 9       35678  168-27 | 137   1235  157   |
| 2356 23567 8     | 12567   34567  16-27  | 13479 12345 14579 |
| 1    23579 2379  | 257     3457  b27     | 8     2345  6     |
*--------------------------------------------------------------*

Sue de Coq: Base Cells r5c56 {26789} Pincer Cells r9c6 {27} + r4c45, r5c5 {5689} => - 7 r1c6, - 5 r4c89, - 568 r6c4, - 27 r78c6;

Code: Select all

*----------------------------------------------------------*
| 239  12379 6     | 4     79-8  19-8  | 5     a138  178   |
| 45   157   147   |d1678  2     3     | 1467   9    147-8 |
| 8    1379  13479 | 167   679   5     | 13467  1346 2     |
|------------------+-------------------+-------------------|
| 7    13689 139   |c568   5689  4     | 2     b168  189   |
| 2469 12689 5     | 3     689   2689  | 1469   7    1489  |
| 2469 2689  249   | 27    1     26789 | 469    5    3     |
|------------------+-------------------+-------------------|
| 2356 4     237   | 9     35678 168   | 137    123  157   |
| 2356 23567 8     | 12567 34567 16    | 13479  1234 14579 |
| 1    23579 2379  | 257   3457  27    | 8      234  6     |
*----------------------------------------------------------*

Skyscraper : (8) r1c8 = r4c8 - r4c4 = (8) r2c4 => - 8 r1c56, r2c9; btte

Leren

[denis_berthier]

.

Code: Select all

Resolution state after Singles and whips[1]:
   +-------------------+-------------------+-------------------+
   ! 239   12379 6     ! 4     789   1789  ! 5     138   178   !
   ! 45    157   147   ! 1678  2     3     ! 1467  9     1478  !
   ! 8     1379  13479 ! 167   679   5     ! 13467 1346  2     !
   +-------------------+-------------------+-------------------+
   ! 7     13689 139   ! 568   5689  4     ! 2     1568  1589  !
   ! 2469  12689 5     ! 3     689   2689  ! 1469  7     1489  !
   ! 2469  2689  249   ! 25678 1     26789 ! 469   4568  3     !
   +-------------------+-------------------+-------------------+
   ! 2356  4     237   ! 9     35678 12678 ! 137   1235  157   !
   ! 2356  23567 8     ! 12567 34567 1267  ! 13479 12345 14579 !
   ! 1     23579 2379  ! 257   3457  27    ! 8     2345  6     !
   +-------------------+-------------------+-------------------+
227 candidates.

z-chain[3]: c4n2{r9 r6} - r6n7{c4 c6} - r9c6{n7 .} ==> r7c6≠2, r8c6≠2
t-whip[3]: r9c6{n7 n2} - c4n2{r9 r6} - r6n7{c4 .} ==> r1c6≠7, r7c6≠7, r8c6≠7
t-whip[3]: r6n7{c4 c6} - r9c6{n7 n2} - c4n2{r9 .} ==> r6c4≠5, r6c4≠6, r6c4≠8
end in S2.

[AnotherLife]

The score given by SE may be too high.

Hi YZF,
I am glad that finally you got interested in the shortbacks of the current rating system. Actually, this puzzle is solvable by an AIC with a group and a basic AIC, and such cases are not rare among the SE 8.3 puzzles.

Code: Select all

.--------------------.------------------------.---------------------.
| 239   12379  6     | 4        789    189-7  | 5      138    178   |
| 45    157    147   | 1678     2      3      | 1467   9      1478  |
| 8     1379   13479 | 167      679    5      | 13467  1346   2     |
:--------------------+------------------------+---------------------:
| 7     13689  139   | 568      5689   4      | 2      1568   1589  |
| 2469  12689  5     | 3        689    2689   | 1469   7      1489  |
| 2469  2689   249   | a27-568  1      d26789 | 469    4568   3     |
:--------------------+------------------------+---------------------:
| 2356  4      237   | 9        35678  168-27 | 137    1235   157   |
| 2356  23567  8     | b12567   34567  16-27  | 13479  12345  14579 |
| 1     23579  2379  | b257     3457   c27    | 8      2345   6     |
'--------------------'------------------------'---------------------'


1. AIC with a group (continuos loop): (2)r6c4 = r89c4 - (2=7)r9c6 - r6c6 = (7-2)r6c4 => -568 r6c4, -27 r78c6, -7 r1c6; 1 single

Code: Select all

.--------------------.---------------------.--------------------.
| 239   12379  6     | 4      789    189   | 5      c138  178   |
| 45    157    147   | a1678  2      3     | 1467   9     b1478 |
| 8     1379   13479 | 167    679    5     | 13467  1346  2     |
:--------------------+---------------------+--------------------:
| 7     13689  139   | 56-8   5689   4     | 2      d168  189   |
| 2469  12689  5     | 3      689    2689  | 1469   7     1489  |
| 2469  2689   249   | 27     1      26789 | 469    5     3     |
:--------------------+---------------------+--------------------:
| 2356  4      237   | 9      35678  168   | 137    123   157   |
| 2356  23567  8     | 12567  34567  16    | 13479  1234  14579 |
| 1     23579  2379  | 257    3457   27    | 8      234   6     |
'--------------------'---------------------'--------------------'


2. Basic AIC (2-string kite): (8)r2c4 = r2c9 - r1c8 = r4c8 => -8 r4c4; lste

Now I am analyzing Denis Berthier's cbg-000 collection of puzzles, and I got the following results about the SE 8.4 puzzles: GC-class - 29,2%, ALSC-class - 45,5%, FC-class - 25,3%.
GC-class - all the puzzles solvable by the standard patterns (SP), AICs with or without groups;
ALSC-class - all the above and all the puzzles solvable by AICs with almost locked sets;
FC-class - all the above and all the puzzles solvable by forcing chains (krakens). I mean that these forcing chains can contain group- and ALS-nodes.

I will publish my statistics later on.
---
EDIT: corrected a typo.

[denis_berthier]

Now I am analyzing Denis Berthier's cbg-000 collection of puzzles, and I got the following results about the SE 8.4 puzzles: GC-class - 29,2%, ALSC-class - 45,5%, FC-class - 25,3%.
GC-class - all the puzzles solvable by the standard patterns (SP), AICs with or without groups;
ALSC-class - all the above and all the puzzles solvable by AICs with almost locked sets;
FC-class - all the above and all the puzzles solvable by forcing chains (krakens). I mean that these forcing chains can contain group- and ALS-nodes

As you claim to do all this manually and there are 293 puzzles with SER 8.4, I doubt we'll ever get anything that could be called statistics.
Not mentioning te size of your sample doesn't make you look very serious about it.

There's nothing as "standard patterns" and this pulls all the analysis down, as nobody knows what you're talking about.

One more absurdity: you want to classify puzzles with a fixed SER (i.e. involving chains of bounded lengths) according to rules with unrestricted length. You've obviously never thought seriously about rating.

If the above "results" were correct, ALS-chains of unrestricted length + all the other undefined patterns would be able to solve only 45% of the puzzles with SER 8.4. That would make them a very weak pattern.

I know they are much less powerful than whips, but I doubt they are so bad.

For serious results, we'll have to wait for yzf...'s implementation of als-chains.

[AnotherLife]

As you claim to do all this manually and there are 293 puzzles with SER 8.4, I doubt we'll ever get anything that could be called statistics.
Not mentioning te size of your sample doesn't make you look very serious about it.

I considered the first 10, 318 puzzles of your collection, and there are 154 puzzles with SER 8.4.

There's nothing as "standard patterns" and this pulls all the analysis down, as nobody knows what you're talking about.

I can specify them. They are conventional methods widely used in practice.

One more absurdity: you want to classify puzzles with a fixed SER (i.e. involving chains of bounded lengths) according to rules with unrestricted length. You've obviously never thought seriously about rating.

If the above "results" were correct, ALS-chains of unrestricted length + all the other undefined patterns would be able to solve only 45% of the puzzles with SER 8.4. That would make them a very weak pattern.

No, I just wanted to show that only 25,3% of SER 8.4 puzzles need forcing chains, and the rest 74,7% are solvable by AICs with groups and ALS's (29,2% - AICs with groups suffice, 45,5% - AICs with ALS's are required). You got me wrong.

[denis_berthier]

I just wanted to show that only 25,3% of SER 8.4 puzzles need forcing chains, and the rest 74,7% are solvable by AICs with groups and ALS's (29,2% - AICs with groups suffice, 45,5% - AICs with ALS's are required). You got me wrong.

But this result is trivially wrong: 0% need forcing chains; all can be solved with short whips.
Whichever way you turn it, it would make the als-chains appear very weak. But your results are not reliable.

[AnotherLife]

But your results are not reliable.

What do you mean?

[yzfwsf]

But this result is trivially wrong: 0% need forcing chains; all can be solved with short whips.

Ordinary single chain, when looking for the possible next node, only consider the influence of the parent node, and your chain will add the influence of the grandparent node and the distant ancestor node. In SE, the kind of influence of adding ancestor nodes is called dynamic chain, which is classified as more complicated than Forcing chain.

YZF

[denis_berthier]

But this result is trivially wrong: 0% need forcing chains; all can be solved with short whips.

Ordinary single chain, when looking for the possible next node, only consider the influence of the parent node, and your chain will add the influence of the grandparent node and the distant ancestor node. In SE, the kind of influence of adding ancestor nodes is called dynamic chain, which is classified as more complicated than Forcing chain.

Thanks. I knew the rules in SE much before you ever heard of Sudoku.
Dynamic chains are not more complicated than longer forcing chains. The SE classification of chains is largely arbitrary, but at least it avoids the beginner's error of putting all the chains of some kind before other types of chains. Therefore, I was perfecfly correct in saying that the claim with the word "need" is false.
All this is based on vague definitions and a total lack of systematic computations. If you ever come to program als-chains and you're able to use short ones before longer ones, I'm curious to see the whole classification of cbg-000 wrt als-chains. That'd be some real statistical result. All the rest is empty talks.