Rating Assassins

Our weekly <a href="http://www.sudocue.net/weeklykiller.php">Killer Sudokus</a> should not be taken too lightly. Don't turn your back on them.
Andrew
Grandmaster
Grandmaster
Posts: 300
Joined: Fri Aug 11, 2006 4:48 am
Location: Lethbridge, Alberta

Post by Andrew »

Andrew wrote:... or a lot of very difficult work. Am I wrong to include the second part of that?
Para wrote:They always have something extra. ... it could be some really heavy combination analysis, really hard use of 45-tests or a very difficult single elimination that is needed to solve the puzzle.
Good to see that reply. I was thinking in particular of really heavy combination analysis or really hard use of 45-tests when I made that comment.
Andrew
Grandmaster
Grandmaster
Posts: 300
Joined: Fri Aug 11, 2006 4:48 am
Location: Lethbridge, Alberta

Post by Andrew »

Ed is planning to extend his Rating Sticky to cover Assassins 1 to 49 including, I assume, variants of these puzzles and other earlier forum puzzles.

He has told me that he would welcome estimated ratings for these puzzles, either in this thread or by PM to him, which could be quoted in the Rating Sticky.

As a start, I've just posted a walkthrough and rating for A17V2. This is a puzzle that Ed encouraged me recently to try again.

May I suggest that people who were forum regulars have a look at the posted walkthroughs for the earlier puzzles; you may still have memories of which ones were easy or hard.

I would also encourage newer forum members, plus those who read the forum and/or solve puzzles but haven't yet posted a walkthrough, to have a go at these earlier puzzles and post your opinions.

Earlier in the thread, it has been suggested that A1 should probably be downrated from 0.75 to 0.5. When I first started solving Assassins A13 had just appeared but I decided to start at A1. It only took me a few days to complete the first 12 Assassins, without using elimination solving, so it's reasonable to assume that most of them are likely to be 0.5 rather than 0.75. Then they started getting harder from A13 onward.

Please support Ed and Richard in their rating efforts! If you are not sure of the definitions for the various rating levels, they are given in the first message of the Rating Sticky. Remember that each rating level covers a range of difficulty levels; the ones quoted in the definitions are usually toward the hardest end of each level.
Andrew
Grandmaster
Grandmaster
Posts: 300
Joined: Fri Aug 11, 2006 4:48 am
Location: Lethbridge, Alberta

Post by Andrew »

As I commented in the Assassin 5 thread, on Mike's definitions we can't rate Assassins lower than 0.75, which was what Caida rated Assassin 5.

Assuming we stick with that definition(*), I would say that most or all of A1 to A17 would be rated 0.75. Even though I still used insertion solving for A18, and possibly a few more beyond that, A18 was distinctly harder than the earlier ones so from memory I'll rate that at 1.0.

(*) At this stage I think the rating definitions are probably best left as they are. There may be some puzzles that need their ratings adjusting, I'll leave that to Ed and Richard, but we shouldn't be changing the definitions now except possibly to indicate whether the "benchmark" puzzles are at the top, centre or bottom of each rating range.

Maybe it's a bit of a generalisation but I would suggest that any Assassin or forum puzzle that needs elimination solving should be rated at least 1.0.

The next big jump in difficulty was Assassin 24. I think Ed said that SS came in with a relatively low rating for that one but I would imagine most human solvers would rate it a bit higher than SS did; I certainly would.
sudokuEd
Grandmaster
Grandmaster
Posts: 257
Joined: Mon Jun 19, 2006 11:06 am
Location: Sydney Australia

Post by sudokuEd »

Andrew wrote:Mike's definitions we can't rate Assassins lower than 0.75, which was what Caida rated Assassin 5
True - though it is working much better for SudokuSolver to score Assassin 1 at closer to .50 (see my previous update post).

Unfortunately, I can't give any definite scores yet. Richard and I are still doing the final, final, final tweaking of what scoring we think works best. It will blow you away when you see all the changes to SudokuSolver that Richard has made (Almost Locked Cages/Hidden Killer Mutuals/Cage Overlap etc, etc). A very nice Chrissie present =P~ . As a result, the scoring for many of the former "problem" puzzles is looking much better.

Lets continue this discussion about what each category means when we all have the next lot of SSscores to analyze - and have the solver to play with. Ho Ho HO!!

Cheers
Ed
Andrew
Grandmaster
Grandmaster
Posts: 300
Joined: Fri Aug 11, 2006 4:48 am
Location: Lethbridge, Alberta

Post by Andrew »

Andrew wrote:Mike's definitions we can't rate Assassins lower than 0.75, which was what Caida rated Assassin 5
sudokuEd wrote:True - though it is working much better for SudokuSolver to score Assassin 1 at closer to .50 (see my previous update post).
It might be better to keep Assassin 1 rated as an easy 0.75 unless there's evidence that it was significantly easier than A2 to A12. If you downgrade A1 to 0.5 then you might have to also downgrade what is currently defined as 0.5.
Andrew wrote:When I first started solving Assassins I found them significantly harder than the daily Diabolicals on www.sudoku.org.uk which I think are about the same level as The Times Deadly that Mike defined as being 0.5.
Andrew wrote:At this stage I think the rating definitions are probably best left as they are. There may be some puzzles that need their ratings adjusting, I'll leave that to Ed and Richard, but we shouldn't be changing the definitions now except possibly to indicate whether the "benchmark" puzzles are at the top, centre or bottom of each rating range..
In view of Ed's comment maybe 0.75, or possibly the lower end of 0.75, might need to be redefined as one of the other very early Assassins if it is thought that A1 was significantly easier.
sudokuEd wrote:Lets continue this discussion about what each category means when we all have the next lot of SSscores to analyze ...
Fair enough. Hope you didn't mind this small contribution limited to one specific part of the rating range.
Ruud
Site Owner
Site Owner
Posts: 601
Joined: Fri Dec 30, 2005 10:21 pm

Comparing notes

Post by Ruud »

After some heavy modifications to SumoCue, I have compiled a table with the ratings for all (currently 82) original Assassins, with the new SumoCue rating, the (average) human player rating and the rating by Richard's Sudoku Solver.

Since player ratings started with A50, there are only a few post-rated early Assassins. It would be nice if we had at least a player rating for A26, A33, A46 and A48.

There are 6 Assassins which cannot yet be solved by the improved version of SumoCue (this used to be more than 20). Assassin 69 is strange: it has a low player rating, while SMQ cannot solve it and SSRB rates it 2.0.

Code: Select all

Assassin rating table
=====================================================================
A ##  SMQ  Play  SSRB |A ##  SMQ  Play  SSRB |A ##  SMQ  Play  SSRB |
----------------------+----------------------+----------------------+
A..1  3.3  0.75  0.50 |A.31  4.2        1.00 |A.61  3.9  1.25  1.38 |
A..2  3.6  0.75  0.88 |A.32  3.3        1.00 |A.62  4.6  0.75  1.25 |
A..3  3.7        1.38 |A.33  5.1        1.38 |A.63  4.1  0.75  1.13 |
A..4  3.5        1.00 |A.34  4.5        1.50 |A.64  4.4  1.25  1.63 |
A..5  3.6  0.75  0.88 |A.35  3.9        0.88 |A.65  5.0  1.25  1.50 |
A..6  3.6        0.88 |A.36  4.3        1.00 |A.66  3.9  1.0   1.50 |
A..7  3.6        0.88 |A.37  4.0        1.38 |A.67  4.2  0.75  1.50 |
A..8  2.6        0.63 |A.38  4.5        1.38 |A.68  4.4  1.0   1.25 |
A..9  3.6  0.75  0.50 |A.39  3.6        0.75 |A.69  -.-  1.25  2.00 |
A.10  3.6  0.75  0.75 |A.40  3.7        0.88 |A.70  3.7  0.75  1.25 |
A.11  3.7        1.13 |A.41  4.6        1.13 |A.71  -.-  1.75  1.00 |
A.12  3.3        0.88 |A.42  4.0        1.25 |A.72  3.6  1.0   1.00 |
A.13  2.6        0.88 |A.43  4.1        1.25 |A.73  4.8  1.0   1.13 |
A.14  3.1        0.63 |A.44  4.3        1.13 |A.74  4.3  1.0   1.75 |
A.15  3.3        0.75 |A.45  4.6        1.13 |A.75  4.6  1.25  1.63 |
A.16  4.4        1.13 |A.46  -.-        1.50 |A.76  3.7  1.0   1.38 |
A.17  3.1        0.75 |A.47  4.5        1.63 |A.77  3.7  1.0   1.25 |
A.18  3.9        1.50 |A.48  5.5        1.75 |A.78  5.5  1.25  1.38 |
A.19  3.5        1.00 |A.49  4.6        1.75 |A.79  4.2  1.0   1.63 |
A.20  3.3        0.88 |A.50  -.-  1.75  2.25 |A.80  -.-  1.5   1.88 |
A.21  3.5        1.25 |A.51  5.5  1.25  1.63 |A.81  4.2  1.25  1.25 |
A.22  4.2        0.88 |A.52  3.6  1.25  0.88 |A.82  4.6  1.25  1.38 |
A.23  4.0        0.88 |A.53  3.6  1.0   1.50 |
A.24  4.8        1.75 |A.54  4.6  1.25  1.75 |
A.25  3.6        1.38 |A.55  4.2  1.25  1.25 |
A.26  5.0        1.63 |A.56  4.2  1.25  1.63 |
A.27  3.6        1.63 |A.57  4.3  1.0   1.00 |
A.28  3.8        1.50 |A.58  4.8  1.0   1.88 |
A.29  4.2        1.25 |A.59  4.2  1.25  1.25 |
A.30  3.3        1.00 |A.60  -.-  1.5   1.75 |
==============================================
Like Sudoku Explainer, I'm using a rating system based on the most difficult technique required. A Killer solvable with singles only will have a score around 1.0. The average Assassin has a score around 4.0. This gives me a little room below the Assassin level to differentiate between easy and very easy killers.

If the technique has measurable elements, different scores are assigned to each measurement.

Currently, the measurable elements are:
1. Cage size
2. Cage sum distance from minimum/maximum
3. Number of rows/columns/nonets in a 45-test
4. Total border length in a 45-test
5. Total number of innies-outies in a 45-test
6. Naked/hidden subset size (2,3,4)

The SumoCue scores are preliminary and may change if the program is subject to further improvements.

Ruud
mhparker
Grandmaster
Grandmaster
Posts: 345
Joined: Sat Jan 20, 2007 10:47 pm
Location: Germany

Post by mhparker »

Ruud wrote:It would be nice if we had at least a player rating for A26, A33, A46 and A48.
Thanks for the offer, Ruud! :roll:

As you can all see, I've just provided a WT and rating for the A33. Maybe one or more fellow forum members can join me in tackling the others that Ruud mentioned above? I hope so.

P.S.:
Ruud wrote:4. Total border length in a 45-test
BTW, what's that?!
Cheers,
Mike
sudokuEd
Grandmaster
Grandmaster
Posts: 257
Joined: Mon Jun 19, 2006 11:06 am
Location: Sydney Australia

Post by sudokuEd »

Finally - update time!

SudokuSolver (v3) Score (SS(v3)score) tries to:
1) make sure the most notorious puzzles (A1, 50, 60, 60RP-Lite, 71 and all the recent ones) score very close to the Rating AND
2) that no easier puzzles (0.75 & 1.0 & low 1.25) score higher than their rating.

Method
To put it very crudely: uses a sum of the number of steps taken by a fixed routine order with individual weightings for each routine.

Generally, the SS(v3)score achieves its purchase through being equally inefficient at solving a puzzle. However, sometimes it finds an efficient way through a puzzle, so the score given will be (much) lower than humans' experience.

A rule-of-thumb is that any SS(v3)score above 0.97 (see Maverick 3) is going to be a hard killer. I'm finding this out this week while trying to make a uA94. The puzzle gets an SS(v3)score of 1.05 - but it's got me beat!! Hope I can finish it to become uA95!!

As always, your feedback is most appreciated. If you feel these scores are close enough, I'll use them to reorganise the Rating Sticky where the human feedback as quoted on the Sticky allows.

NOTE 1:lowest rotated score used for each puzzle (to minimize the effect of puzzle orientation on SS's solving routine). The original orientation's score may differ.

NOTE 2: variations for A50-59 not included since these don't correlate well and were from before we started consciously thinking about ratings.

NOTE 3: Puzzles that are 1-rating out are noted in the +/- 0.25 column [edit: look for the ones with "OUT" or "JUST in"]. Your reactions to any that are "OUT" or "JUST in" will be really interesting. Those reactions might make there way onto the rating sticky :) .

NOTE 4: We may need to have a discussion about adapting Mike's original rating definitions now that we have some firm numbers to work with.


SS(v3)scores for Rating Sticky Killers

Code: Select all

               |  Sticky  |SS&#40;v3&#41;score| +/- 0.25?|
------------------------------------------------------------
uAssassin 93    |   1.0    |  0.95     |     in

"Maverick" 3    |   1.25   |  0.97     |    OUT   |
"Maverick" 4    |Hard 1.50 |  2.57     |    OUT   |

Assassin 91     |   1.25   |  1.16     |   in

Assassin 90     |Hard 1.75 |  2.29&#40;T&E&#41;|    OUT   |

Assassin 89     |   1.25   |  1.19     |   in
"Bored89-Hard"  |   1.75   |  1.63     |   in
"Bored89-Easy"  |   1.25   |  1.28     |  in

Assassin 88     |Hard 1.25 |  1.52     |    OUT   |

Assassin 87     |Hard 1.0  |  1.10     |

Assassin 86     |  1.25    |  1.01     |  JUST in|

A85 "Original"  |  2.50    |  1.94&#40;T&E&#41;|    OUT   |
A85 "Suitable"  |  1.25    |  1.21     |   in

"Radial"        |  1.25    |  1.22     |   in

Assassin 84     |  1.25    |  1.36     |   in
Assassin 84 V2  |  1.50    |  2.38     |    OUT   |

Assassin 83     |  1.50    |  1.59     |   in

Assassin 82     |  1.25    |  0.97     |    OUT   |
Assassin 82 V2X |  1.50    |  1.46     |   in

Assassin 81     |  1.00    |  1.06     |   in

Assassin 80     |  1.50    |  1.68     |   in

Assassin 79     |  1.00    |  1.02     |   in
A79 "RP"        |  1.50    |  1.45     |   in

Assassin 78     |  1.25    |  1.29     |   in
Assassin 78 V2  |Hard 1.50 |  2.72     |    OUT   |

"Maverick" 2    |  1.75    |  1.89     |   in

Assassin 77     |  1.00    |  1.03     |   in
Assassin 77 V2  |  1.25    |  0.99     |    OUT   |

Assassin 76     |  1.00    |  1.02     |   in
Assassin 76X    |  1.25    |  1.38     |   in

"Diagonal 
surprise"       |  1.25    |  1.12     |   in

Assassin 75     |  1.25    |  1.33     |   in

Assassin 74     |  1.00    |  0.93     |   in
Assassin 74 V2  |  1.25    |  1.25     |    in
A74"Brick Wall" |  2.00    |  1.96     |   in


"Maverick" 1    |  1.75    |  1.59     |   in

Assassin 73     |  1.00    |  1.00     |   in
Assassin 73 V1.5|Hard 1.50 |  1.80     |    OUT   |

Assassin 72     |  1.00    |  1.08     |   in
A72"Old School" |  1.75    |  2.62     |    OUT   |

Concentric 
Squares         |  1.25    |  1.29     |    in

Assassin 71     |  1.75    |  1.72     |    in
A71"Full Border"|  1.50    |  1.11     |    OUT   |
Assassin 71 V1.5|  1.25    |  1.30     |    in

Assassin 70     |  0.75    |  0.95     |  JUST in|
Assassin 70 V2  |  1.75    |  1.57     |    in
Assassin 70 V3  |  1.25    |  1.03     |  JUST in|

Assassin 69     |  1.25    |  1.21     |  in
Assassin 69 V1.5|  1.50    |  1.73     |  JUST in|

Assassin 68     |  1.00    |  1.21     |  JUST in|
Assassin 68 V2  |  1.75    |  1.39     |   OUT    |
Assassin 68 V3  |  3.00    |  5.11&#40;T&E&#41;|   OUT    |
Assassin 68 V1.5|  1.25    |  1.28     |   in

Transformer X   |  1.75    |  1.66     |   in
Trans X Lite    |  1.25    |  1.43     |   in

Vortex X        |  1.75    |  1.77     |    in
Vortex X Lite   |  1.25    |  1.02     |   OUT    |

Assassin 67     |  0.75    |  0.94     |    in

Assassin 66     |  1.0     |  0.99     |    in
Assassin 66 V1.5|  1.5     |  1.71     |  JUST in|

Assassin 65     |  1.25    |  1.08     |   in
Assassin 65 V2  |  1.75    |  2.51     |   OUT    |
Assassin 65 V3  |  1.75    |  1.74     |   in

Assassin 64     |  1.25    |  1.21     |    in
Assassin 64V2   |  1.75    |  1.86     |    in

Assassin 63     |  0.75    |  0.94     |  JUST in|
Assassin 63 V1.5|  1.00    |  0.96     |   in
Assassin 63 V2  |  1.50    |  1.60     |    in

Assassin 62     |  0.75    |  1.0      |    OUT   |
Assassin 62V2   |  2.00    |  3.19     |    OUT   |

Assassin 61     |  1.25    |  1.34     |    in
Assassin 61X    |  1.50    |  1.81     |    OUT   |

Assassin 60     |  1.50    |  1.50     |   in
Assassin 60 "RP"|  2.50    |  2.70&#40;T&E&#41;|   in &#40;edit to correct rating&#41;
A60 "RP-Lite"   |  1.75    |  2.67     |    OUT   |

Assassin 59     |  1.25    |  1.36     |   in

Assassin 58     |  1.0     |  1.04     |   in

Assassin 57     |  1.0     |  0.98     |   in

Assassin 56     |  1.25    |  0.75     |    OUT   |

Assassin 55     |  1.25    |  1.49     |   JUST in|

Assassin 54     |  1.25    |  1.01     |   JUST in|

Assassin 53     |  1.00    |  0.87     |   in

Assassin 52     |  1.25    |  1.05     |  JUST in|

Assassin 51     |  1.25    |  1.02     |  JUST in|

Assassin 50     |  1.75    |  1.75     |   in

A48-Hevvie      |  2.5     |  4.98     |   OUT    |

Assassin 2X     |  1.50    |  1.75     |   OUT    |
Assassin 2X Lite|  1.00    |  1.09     |   in

Assassin 1      |  0.75    |  0.62     |   in
Assassin 1V2    |  1.25    |  1.06     |  JUST in| 
Last edited by sudokuEd on Sat Mar 15, 2008 9:41 am, edited 1 time in total.
mhparker
Grandmaster
Grandmaster
Posts: 345
Joined: Sat Jan 20, 2007 10:47 pm
Location: Germany

Post by mhparker »

Thanks, Ed!
sudokuEd wrote:Finally - update time!
Yippee! :D :bounce:
sudokuEd wrote:NOTE 3: Puzzles that are 1-rating out are noted in the +/- 0.25 column.
What's happened to that last column? Several puzzles are marked as "yes", even though they are way over 0.25 out, whilst others are within +/- 0.25, but are marked as "nearly" or even "no"!

Look forward to going through the list in more detail later.
Cheers,
Mike
sudokuEd
Grandmaster
Grandmaster
Posts: 257
Joined: Mon Jun 19, 2006 11:06 am
Location: Sydney Australia

Post by sudokuEd »

mhparker wrote:What's happened to that last column?
Hope it's clearer now. I couldn't put smiley emoticons on all the "in" puzzles - but they are with-in 1 rating of the sticky.

Look for the ones that say "OUT" - they are more than 1 rating +/- from the Sticky. "JUST in" are borderline - so think about them as well.

At least I got (pretty) straight columns this time - thanks for the tip Nasenbaer :idea: .

Any more tips for presentation let me know by PM and I'll do my best!!

Thanks Mike.
Ed
Afmob
Expert
Expert
Posts: 103
Joined: Sat Sep 22, 2007 5:36 pm
Location: MV, Germany

Post by Afmob »

I'm not so sure it would be such a good idea to include the SS scores in the rating sticky since:

1) They depend on the version of SS, so an update to SS means an updates to the scores and the sticky.
2) There are too many cases where the rating is "out".
3) You don't need to look for SS Scores here since you use SudokuSolver on your own where as the human ratings cannot be replaced (yet) by a software (2) and can only be found in the sticky.

I think especially hard Assassins (1.75+) are often "out" since it's tough to judge how difficult a chain/heavy combo analysis is and whether they are T&E or not. Using Mike's approach only chains which are AIC would be acceptable (so rating <= 2.0 if they are not too long) otherwise they are T&E and so the rating must be 2.5+.
And since most puzzles aren't of such difficulty we have only few examples and wts for such monsters, which is quite understandable if you have ever tackled one of those Killers. So the main focus for adapting SS Scores should be Assassins of rating <(=?) 1.75.
Andrew
Grandmaster
Grandmaster
Posts: 300
Joined: Fri Aug 11, 2006 4:48 am
Location: Lethbridge, Alberta

Post by Andrew »

Great work Ed and Richard! =D>

Some very good points by Afmob. I agree that SS scores shouldn't go in the rating sticky because they depend on the version of SS. BTW the original rating sticky for A50 to A74 said that a few puzzles couldn't be solved by SS, stating how many steps were done. I wonder if those entries are still valid. I'm not suggesting that statements of how many steps were done should be removed, just updated if appropriate since we know that SS has been significantly improved since that sticky was first posted.

However there is still a valid place for SS scores in messages like Ed's one in this thread, where they are given in the context of a particular version of SS.

Is it really true that the only acceptable chains are AIC and that any others are T&E? Mike has made the point in previous discussions that AIC don't alter the grid state whereas some other steps that start with a "what if" assumption do change the state as they progress through a chain. That is clearly a valid distinction. All chains, whether AIC, contradiction chains, forcing chains or pure T&E must start with a "what if" assumption.

Then there is methodical combination/permutation analysis. Sometimes in my walkthroughs I present this as a series of sub-steps so it may at first sight appear to look like T&E. Others will just give a long string of combinations and permutations which don't give that same impression. How it contributes to the rating will clearly depend on the level of the analysis. In most cases a puzzle needing methodical combination/permutation analysis will be at least a Hard 1.25 and more likely a 1.5; some will rate higher than that if really heavy analysis or repeated analysis is required.

Ed asked us to comment on ones that are Just In or are Out. Of course that depends on one's interpretation of the range covered by each rating level. I must admit I was surprised when, in discussion with Ed, I learned that he considers the rating 1.25 to cover scores from 1.25 to 1.49; I subsequently learned that Para has the same interpretation. I had interpreted 1.25 to cover roughly 1.15 to 1.35 and still take that view. On that basis some of the ones that Ed scores In are, in my view out, and vice versa. The difference of interpretation only matters when comparing human ratings with SS scores which clearly must follow Ed's interpretation. Among humans the difference doesn't matter. If I rate a puzzle as High 1.25 meaning 1.35 and Ed gives the same rating meaning 1.45, we both know what we mean by a Hard 1.25; one that isn't quite hard enough for a 1.5 rating.

Even on Ed's interpretation there are several that are Just In, ranging from 0.21 to 0.24 above the Sticky rating but there is one that is 0.25 above so is listed as OUT.

Maybe it would be more realistic, IMHO, to consider that SS scores ought only to be quoted to the nearest 0.05, particularly since Ed mentions that puzzle rotation is used with sometimes different scores for the various orientations. I realise there would be a downside to this since any that are currently Just In at 0.23 and 0.24 would become OUT at 0.25.

There were a few cases where I was amazed at the differences.

A71 (Full Border) was one that Mike and I did as a "tag", mostly Mike with a few steps from me in the middle. While it was a relatively easy puzzle for a "tag" solution, I'm completely surprised that SS rated it as low as 1.11.

Mav 4 is scored at 2.57 by SS. Having solved that puzzle, although I haven't yet gone through posted walkthroughs or posted my one, I don't see how it can be that high. It took a bit of methodical combination/permutation analysis but not particularly heavy work.

A85 (Original Version), in which I participated a little in the "tag" solution was a real brute. I don't understand how SS rates it as 1.94 but at the same time T&E is stated. If a software solver needs to use T&E how can the rating be that low?

Discrepancies for the very highest rated ones are hardly surprising. I think Afmob was suggesting that we shouldn't be concerned about that, which I agree with.

Finally Ed's NOTE 4. "We may need to have a discussion about adapting Mike's original rating definitions now that we have some firm numbers to work with". Sorry but I must disagree with that. After all the discussions and work that has been put in, surely it's too late for that unless some minor clarification changes are being suggested. Some existing ratings for individual puzzles in the sticky may need to be changed; that suggestion has been around for some time. I know for example that several people think that Mav 1 should be downgraded from 1.75 to 1.5. If that happens I'll accept the majority view but it won't change my personal view of that puzzle :wink: .
sudokuEd
Grandmaster
Grandmaster
Posts: 257
Joined: Mon Jun 19, 2006 11:06 am
Location: Sydney Australia

Post by sudokuEd »

Andrew wrote:Great work Ed and Richard! =D>
Thanks!
Andrew wrote:I had interpreted 1.25 to cover roughly 1.15 to 1.35 and still take that view
At the risk of causing confusion, I've adapted the list - rounded to .05, and WITHIN 0.15 (not 0.25)and without any use of UR which makes a big difference to A71 Full Border (and A50 [edit:& A74 BrickWall]).

Trans X Lite is the one I'm least happy about getting into the 1.50 bracket. Many of the others that are HIGH tend to reflect (at least some) of the actual experience of solvers where there was a range of opinions.
Andrew wrote:A71 (Full Border) .. I'm completely surprised that SS rated it as low as 1.11
No UR makes a huge difference to this one.
Andrew wrote:Mav 4 is scored at 2.57 by SS
Haven't looked into this one yet but sounds like SS needs to be taught something new.
Andrew wrote:A85 (Original Version), in which I participated a little in the "tag" solution was a real brute. I don't understand how SS rates it as 1.94 but at the same time T&E is stated. If a software solver needs to use T&E how can the rating be that low?
This one is an anomaly [edit:fluke is probably a better word]. Shutting down all the hidden & locked cage options in SS (to help with many) means that it finds a very efficient solution for this one using "45 Insane i/o". We might need to bump up the weighting given to that routine. Here are the stats.

BTW - I really like many of the "simple Bowman Bingo Lite" that SS finds. Feels like the sort that we often use.
Assassin 85 Original

Code: Select all

40	Naked Singles
10	Naked Pairs
2	Naked Triples
9	Hidden Singles
5	Locked Candidates &#40;Box/Line&#41;
1	Cage Blockers
2	Cage Placement	3
1	Cage Placement	2
3	Cage Placement	1
6	Cage Placement
7	Cage Combinations
1	Cage Placement Extended	1
1	Cage Placement Extended
1	45 Rule Two Innies/Outies	2
2	45 Rule Single House	4
4	45 Rule Extended	5
3	45 Rule Extreme	5
1	Common Peer Elimination Extended	4
60	Cage Cleanup
1	45 Rule Multiple Houses Innies&Outies	5
1	45 Rule Extended Innies&Outies	4
1	45 Rule Insane Innies&Outies	7
1	45 Rule Insane Innies&Outies	6
2	45 Rule Insane Innies&Outies	5
1	Bowmans Bingo Simple &#40;Lite&#41;	1
4	Locked Candidates &#40;House/Cage&#41;
6	Forced Cage Candidates
1	Cage Blockers Insane
1	Forced Cage Candidates - Complex
5	45 Rule Simple Single House	3
1	45 Rule Simple Single House Innies&Outies	3
Total solving time &#40;seconds&#41;&#58;		9.22
Calculated score&#58;		1.94

**** Uses Simple Trial & Error ****
Andrew wrote:Discrepancies for the very highest rated ones are hardly surprising. I think Afmob was suggesting that we shouldn't be concerned about that, which I agree with
I've left these ones alone for this list.

SS(v3)scores for Rating Sticky Killers NOTE:within 0.15

Code: Select all

               |  Sticky  |SS&#40;v3&#41;score| result within 0.15?|
------------------------------------------------------------
uAssassin 93    |   1.0    |  0.95     |     

"Maverick" 3    |   1.25   |  0.95     |    LOW   |
"Maverick" 4    |Hard 1.50 |  2.60     |    HIGH  |

Assassin 91     |   1.25   |  1.15     |   

Assassin 90     |Hard 1.75 |  2.30&#40;T&E&#41;|          

Assassin 89     |   1.25   |  1.20     |   
"Bored89-Hard"  |   1.75   |  1.65     |   
"Bored89-Easy"  |   1.25   |  1.30     |  

Assassin 88     |Hard 1.25 |  1.50     |    HIGH  |

Assassin 87     |Hard 1.0  |  1.10No UR|   

Assassin 86     |  1.25    |  1.00     |    LOW   |

A85 "Original"  |  2.50    |  1.95&#40;T&E&#41;|          
A85 "Suitable"  |  1.25    |  1.20     |   

"Radial"        |  1.25    |  1.20     |   

Assassin 84     |  1.25    |  1.35     |   
Assassin 84 V2  |  1.50    |  2.40     |   HIGH   |

Assassin 83     |  1.50    |  1.60     |   

Assassin 82     |  1.25    |  0.95     |   LOW    |
Assassin 82 V2X |  1.50    |  1.45     |   

Assassin 81     |  1.00    |  1.05     |   

Assassin 80     |  1.50    |  1.70     |   HIGH   |

Assassin 79     |  1.00    |  1.00     |   
A79 "RP"        |  1.50    |  1.45No UR|   

Assassin 78     |  1.25    |  1.30     |   
Assassin 78 V2  |Hard 1.50 |  2.70     |   HIGH   |

"Maverick" 2    |  1.75    |  1.90     |   

Assassin 77     |  1.00    |  1.05     |   
Assassin 77 V2  |  1.25    |  1.00     |   LOW    |

Assassin 76     |  1.00    |  1.00     |   
Assassin 76X    |  1.25    |  1.40     |     

"Diagonal 
surprise"       |  1.25    |  1.10     |   

Assassin 75     |  1.25    |  1.35     |   

Assassin 74     |  1.00    |  0.95     |   
Assassin 74 V2  |  1.25    |  1.25     |    
A74"Brick Wall" |  2.00    |  2.90No UR|   


"Maverick" 1    |  1.75    |  1.60     |   

Assassin 73     |  1.00    |  1.00     |   
Assassin 73 V1.5|Hard 1.50 |  1.80     |   HIGH   |

Assassin 72     |  1.00    |  1.10     |   
A72"Old School" |  1.75    |  2.62     |          

Concentric 
Squares         |  1.25    |  1.30     |    

Assassin 71     |  1.75    |  1.70     |    
A71"Full Border"|  1.50    |  1.60No UR|    
Assassin 71 V1.5|  1.25    |  1.30     |    

Assassin 70     |  0.75    |  0.95     |    HIGH  |
Assassin 70 V2  |  1.75    |  1.55     |    LOW   |
Assassin 70 V3  |  1.25    |  1.05     |    LOW   |

Assassin 69     |  1.25    |  1.20     |    
Assassin 69 V1.5|  1.50    |  1.75     |    HIGH  |

Assassin 68     |  1.00    |  1.20     |    HIGH  |
Assassin 68 V2  |  1.75    |  1.40     |    LOW   |
Assassin 68 V3  |  3.00    |  5.11&#40;T&E&#41;|         
Assassin 68 V1.5|  1.25    |  1.30     |   

Transformer X   |  1.75    |  1.65No UR|   
Trans X Lite    |  1.25    |  1.45     |    HIGH  |

Vortex X        |  1.75    |  1.75     |    
Vortex X Lite   |  1.25    |  1.00     |    LOW   |

Assassin 67     |  0.75    |  0.95     |    HIGH  |

Assassin 66     |  1.00    |  1.00     |    
Assassin 66 V1.5|  1.50    |  1.70     |    HIGH  |

Assassin 65     |  1.25    |  1.10     |    

Assassin 65 V2  |  1.75    |  2.50     |         
Assassin 65 V3  |  1.75    |  1.75     |   

Assassin 64     |  1.25    |  1.20     |    
Assassin 64V2   |  1.75    |  1.85     |    

Assassin 63     |  0.75    |  0.95     |    HIGH  |
Assassin 63 V1.5|  1.00    |  0.95     |    
Assassin 63 V2  |  1.50    |  1.60     |    

Assassin 62     |  0.75    |  1.0      |    HIGH  |
Assassin 62V2   |  2.00    |  3.20     |          

Assassin 61     |  1.25    |  1.3oNo UR|    
Assassin 61X    |  1.50    |  1.80     |    HIGH  |

Assassin 60     |  1.50    |  1.50     |   
Assassin 60 "RP"|  2.50    |  2.70&#40;T&E&#41;|   
A60 "RP-Lite"   |  1.75    |  2.65     |          

Assassin 59     |  1.25    |  1.35     |   

Assassin 58     |  1.0     |  1.05     |   

Assassin 57     |  1.0     |  1.00     |   

Assassin 56     |  1.25    |  0.75     |    LOW   |

Assassin 55     |  1.25    |  1.45No UR|    HIGH  |

Assassin 54     |  1.25    |  1.00     |    LOW   |

Assassin 53     |  1.00    |  0.85     |    

Assassin 52     |  1.25    |  1.05     |    LOW   |

Assassin 51     |  1.25    |  1.00     |    LOW   |

Assassin 50     |  1.75    |  2.50No UR|   

A48-Hevvie      |  2.5     |  5.00     |          

Assassin 2X     |  1.50    |  1.75     |   HIGH   |
Assassin 2X Lite|  1.00    |  1.10     |   

Assassin 1      |  0.75    |  0.60     |   
Assassin 1V2    |  1.25    |  1.05     |   LOW    | 
Last edited by sudokuEd on Wed Mar 19, 2008 11:20 pm, edited 1 time in total.
mhparker
Grandmaster
Grandmaster
Posts: 345
Joined: Sat Jan 20, 2007 10:47 pm
Location: Germany

Post by mhparker »

Andrew wrote:I agree that SS scores shouldn't go in the rating sticky because they depend on the version of SS.
I would also go along with this view. Seems like far too much maintenance overhead for poor old Ed to me, who would not only have to append the latest SScore to the list each time a new version of SS is released, but would also have to re-categorize and re-sort the list accordingly each time.
Andrew wrote:Is it really true that the only acceptable chains are AIC and that any others are T&E?
Probably not, but it is my experience that most people underestimate the complexity of the chains they use.
Andrew wrote:All chains, whether AIC, contradiction chains, forcing chains or pure T&E must start with a "what if" assumption.
Actually, AICs should strictly speaking be found by locating and linking up the (relatively few) strong links in the grid, rather than by assuming that a particular cell (which?) contains a particular digit (which?)...
Andrew wrote:I must admit I was surprised when, in discussion with Ed, I learned that he considers the rating 1.25 to cover scores from 1.25 to 1.49; I subsequently learned that Para has the same interpretation. I had interpreted 1.25 to cover roughly 1.15 to 1.35 and still take that view.
This is also the approach I take when posting a puzzle. If I think a puzzle is maybe a "1.4", I usually just post it as a "1.5" and be done with it. In this case, I just see the rating as being a rough guide.
Cheers,
Mike
Andrew
Grandmaster
Grandmaster
Posts: 300
Joined: Fri Aug 11, 2006 4:48 am
Location: Lethbridge, Alberta

Post by Andrew »

Thanks Ed for the update to the SS scores, rounding them to 0.05. It's interesting that you have also chosen to narrow down your comparison from +/- 0.25 to +/- 0.15. It's good that you have the confidence to do that. :D

It's interesting how removing UR from the SS score has little effect on several puzzles, brings A71 "Full Border" almost spot on but massively increases the SS score for A74 Brick Wall.

I won't re-open any discussion about UR, which some find acceptable and others don't. However I would suggest that if UR is to be used it probably ought to be considered more difficult than in a 1.11 rating, which is what A71 "Full Border" scored with UR. Of course I may be biased in saying that because I'm one who refuses to use it; it may be fairly easy to spot for those who use it.
Mike wrote:Actually, AICs should strictly speaking be found by locating and linking up the (relatively few) strong links in the grid, rather than by assuming that a particular cell (which?) contains a particular digit (which?)...
A very good point Mike. If people use it that way then it certainly reduces the "what if" aspect. Of course there may well still be the "what if" aspect present when you re-write as AICs some steps that others have given as contradiction moves.

I'll have to try to remember that point about looking for the relatively few strong links if/when I learn AICs. I may take Andrew Stuart's book, which has been on a bookcase untouched for quite a long time, with me when we go on holiday in June. However I'm not sure how much I'll be able to learn about advanced techniques if I've only got the book without a computer to try out things on an Excel worksheet.
Post Reply