The New Sudoku Players' Forum

by **gsf** » Sun Apr 16, 2006 8:23 am

tarek wrote:
Code: Select all
Rank Puzzle Compiler Score Variety 100 Puzzle#001 (tarek) 3 1

thanks
I have #001 with 1 xwing (3) and 1 naked pair (1) => 4
how do you get 3?

by **tarek** » Sun Apr 16, 2006 8:54 am

neither of the naked doubles (in column 5 or box 8) are needed if you spot the xwings first, I programmed my solver to spot doubles before xwings & therefore the score as you said should be 4, some members spot the xwings before the doubles, so another routine with the xwings just after box line interactions was introduced for this purpose only... if both routines spotted a double then it registers if one spots a double but not the other then (as in #001) it doesn't.....

tarek

by **gsf** » Mon Apr 17, 2006 4:13 pm

tarek wrote:2. Hidden Triple*6+XWing*3+Naked triple*2+Hidden double*2+Naked double is the general formula.

still having trouble reproducing results
(although I uncovered some naked/hidden counting bugs of my own)
could you outline how #074 gets a rating of 15?
thanks

by **tarek** » Mon Apr 17, 2006 6:56 pm

The scoring system identifies (almost) the techniques the are needed to solve....there maybe extra depending on how you solve it.... but the techniques scored are the ones that show up no matter how you solve it...

in the #074 case (a brilliant puzzle by the way).....there is 1 Hidden triple...1 X-wing.....1 Naked triple....1 Hidden double ....2 Naked doubles ....scoring 15...

There is more if you're not the x-wing spotter. For the sake of insuring consistancy, if there was a mistake in counting, it would be underscored rather than overscored...... Look at puzzle #100, It scored a 7 although however you try to solve it will score at least a 14....

tarek

by **gsf** » Mon Apr 17, 2006 8:45 pm

tarek wrote:The scoring system identifies (almost) the techniques the are needed to solve....there maybe extra depending on how you solve it.... but the techniques scored are the ones that show up no matter how you solve it...

in the #074 case (a brilliant puzzle by the way).....there is 1 Hidden triple...1 X-wing.....1 Naked triple....1 Hidden double ....2 Naked doubles ....scoring 15...

There is more if your not the x-wing spotter. For the sake of ensuring consistancy, if there was a mistake in counting, it would be underscored rather than overscored...... Look at puzzle #100, It scored a 7 although however you try to solve it will score at least a 14....

the original rating equation and subsequent descriptions are insufficient to repeat the experiment
e.g., what does "if your not the x-wing spotter" mean exactly?
do you have one program that produces the rating or are you combining
info from several sources to determine "the ones that show up no matter how you solve it"?
has anyone else reproduced tarek's numbers using non-tarek code?
I'm interested in general terminology that can be applied to general solvers to reproduce results
i.e., one should not have to re-implement tarek's solver to reproduce the ranking

for example, the inferior thread specifies a technique order and
when a technique produces a placement or elimination the solver
restarts again from the first technique (with special case batching
for naked/hidden single placements)

I realize this is mainly a fun thread and appreciate the work behind compiling the list,
and I'm not advocating re-rating this thread, just reproducable specs for
the next time a thread like this is proposed -- then submitters can
let their generators fly with a reasonable expectation of matching
rankings produced by other solvers

here is my stab at the technique order:
(xx) don't count hidden/naked singles
(T2) naked 2-tuple (double)
(H2) hidden 2-tuple
(W2) x-wing (order 2)
(T3) naked 3-tuple (triple)
(H3) hidden 3-tuple
and tarek's rating equation mapped to the above

Code: Select all: 6*H3 + 2*T3 + 3*W2 + 2*H2 + T2

where, e.g., H3 would be counted at a particular position only if T2..T3 failed to make progress
the rating for puzzle #074 would be 22=(6*1+2*1+3*2+2*3+2)

by **ronk** » Mon Apr 17, 2006 8:55 pm

gsf wrote:here is my stab at the technique order:
(xx) don't count hidden/naked singles
(T2) naked 2-tuple (double)
.............................

With uncounted line/box interaction between those two ???

by **tarek** » Mon Apr 17, 2006 9:08 pm

gsf wrote:I'm not advocating re-rating this thread, just reproducable specs for the next time a thread like this is proposed -- then submitters can let their generators fly with a reasonable expectation of matching
rankings produced by other solvers

Not at all, I myself would like to see some other form of rating...or anyone trying to rate it according to the equation....

I just started the thread, the contributions were from the members, any contributions towards rating would only add to the RICHNESS of this thread.

also as ron just mentioned, box-line interactions did not count into scoring... so 18 needed box line interactions would be as good as 0.

As you know I mentioned earlier that 2 similar tequeniques which can be deduced from the same canididate grid are counted as 1, this has harshly downscored several puzzles (a brilliant puzzle from vidarino had 2 hidden triples, both of them were NEEDED to solve, however both of them were visible from the same candidate grid....that is why they were counted as 1, you may consider this as a weakness & this is were different solvers would score it differently), one of henk's puzzles had 8 x-wings visible at the same step (2 were actually required) but at the end it scored only 1 x-wing, your puzzle #031 scored 3 x-wings (3 x-wings were needed at different steps, the only one in the list that had that

)....

tarek

by **gsf** » Mon Apr 17, 2006 9:25 pm

ronk wrote:
gsf wrote:here is my stab at the technique order:
(xx) don't count hidden/naked singles
(T2) naked 2-tuple (double)
.............................

With uncounted line/box interaction between those two ???

right - I was focusing on the one's tarek counted
other techniques should also be handled: W3 (swordfish) and W4 (jellyfish) and Xn/Yn for X/Y cycles, C* for coloring, etc.

its faily easy to code an expression evaluator on <technique><order> variables (e.g., W2, T3)
we just have to agree on what W2==3 means
(3 separate steps required an x-wing where no other lesser technique made progress)

providing runtime control over technique order is a separate issue
but prompted by this thread I at least split the naked/hidden/wing techniques by size
T2H2W2 (naked/hidden doubles, x-wing)
T3H3W3 (naked/hidden triples, swordfish)
T4H4W4 (naked/hidden quads, jellyfish)

what I'd like to see in a similar thread:
given this technique order T2... and this expression (5*T3...) who can produce a puzzle with the smallest/largest expression value

thanks

by **gsf** » Mon Apr 17, 2006 9:33 pm

tarek wrote:As you know I mentioned earlier that 2 similar tequeniques which can be deduced from the same canididate grid are counted as 1, this has harshly downscored several puzzles (a brilliant puzzle from vidarino had 2 hidden triples, both of them were NEEDED to solve, however both of them were visible from the same candidate grid....that is why they were counted as 1, you may consider this as a weakness & this is were different solvers would score it differently), one of henk's puzzles had 8 x-wings visible at the same step (2 were actually required) but at the end it scored only 1 x-wing, your puzzle #031 scored 3 x-wings (3 x-wings were needed at different steps, the only one in the list that had that )....

thanks
one of my problems may have been "similar"
for #074 one step had a hidden double followed by a hidden triple
how did you count that?

by **tarek** » Mon Apr 17, 2006 10:08 pm

If you spot the hidden double first.....This is what you see

Code: Select all: *--------------------------------------------------------------------------* | 3458 35789 3489 | 1 458 6 | 5789 2789 2789 | | 4568 578 1 | 245 9 2458 | 3 78 678 | | 568 2 689 | 7 58 3 | 5689 4 1 | |------------------------+------------------------+------------------------| | 7 368 2 | 69 1368 89 | 4 189 5 | | 1 4 38 | 259 3578 2589 | 789 6 789 | | 9 68 5 | 46 14678 48 | 2 178 3 | |------------------------+------------------------+------------------------| | 2346 1 3469 | 8 46 7 | 69 5 2469 | | 4568 58 7 | 4569 2 459 | 1 3 468 | | 24568 589 4689 | 3 456 1 | 6789 2789 246789 | *--------------------------------------------------------------------------* r6c5 Must only have 17 as valid Candidates (17 is a Hidden Double in Row 6) r6c8 Must only have 17 as valid Candidates (17 is a Hidden Double in Row 6) *--------------------------------------------------------------------------* | 3458 35789 3489 | 1 458 6 | 5789 2789 2789 | | 4568 578 1 | 245 9 2458 | 3 78 678 | | 568 2 689 | 7 58 3 | 5689 4 1 | |------------------------+------------------------+------------------------| | 7 368 2 | 69 1368 89 | 4 189 5 | | 1 4 38 | 259 3578 2589 | 789 6 789 | | 9 68 5 | 46 17 48 | 2 17 3 | |------------------------+------------------------+------------------------| | 2346 1 3469 | 8 46 7 | 69 5 2469 | | 4568 58 7 | 4569 2 459 | 1 3 468 | | 24568 589 4689 | 3 456 1 | 6789 2789 246789 | *--------------------------------------------------------------------------* r4c5 Must only have 13 as valid Candidates (137 is a Hidden Triple in Column 5) r5c5 Must only have 37 as valid Candidates (137 is a Hidden Triple in Column 5) r4c5 Must only have 13 as valid Candidates (137 is a Hidden Triple in Box 5) r5c5 Must only have 37 as valid Candidates (137 is a Hidden Triple in Box 5)

but look at from this way:

Code: Select all: *--------------------------------------------------------------------------* | 3458 35789 3489 | 1 458 6 | 5789 2789 2789 | | 4568 578 1 | 245 9 2458 | 3 78 678 | | 568 2 689 | 7 58 3 | 5689 4 1 | |------------------------+------------------------+------------------------| | 7 368 2 | 69 1368 89 | 4 189 5 | | 1 4 38 | 259 3578 2589 | 789 6 789 | | 9 68 5 | 46 14678 48 | 2 178 3 | |------------------------+------------------------+------------------------| | 2346 1 3469 | 8 46 7 | 69 5 2469 | | 4568 58 7 | 4569 2 459 | 1 3 468 | | 24568 589 4689 | 3 456 1 | 6789 2789 246789 | *--------------------------------------------------------------------------* r6c5 Must only have 17 as valid Candidates (468 is a Naked Triple in Row 6) r6c8 Must only have 17 as valid Candidates (468 is a Naked Triple in Row 6) *--------------------------------------------------------------------------* | 3458 35789 3489 | 1 458 6 | 5789 2789 2789 | | 4568 578 1 | 245 9 2458 | 3 78 678 | | 568 2 689 | 7 58 3 | 5689 4 1 | |------------------------+------------------------+------------------------| | 7 368 2 | 69 1368 89 | 4 189 5 | | 1 4 38 | 259 3578 2589 | 789 6 789 | | 9 68 5 | 46 17 48 | 2 17 3 | |------------------------+------------------------+------------------------| | 2346 1 3469 | 8 46 7 | 69 5 2469 | | 4568 58 7 | 4569 2 459 | 1 3 468 | | 24568 589 4689 | 3 456 1 | 6789 2789 246789 | *--------------------------------------------------------------------------* r4c5 Must only have 13 as valid Candidates (137 is a Hidden Triple in Column 5) r5c5 Must only have 37 as valid Candidates (137 is a Hidden Triple in Column 5) r4c5 Must only have 13 as valid Candidates (137 is a Hidden Triple in Box 5) r5c5 Must only have 37 as valid Candidates (137 is a Hidden Triple in Box 5)

which is more difficult how do you score it???

I only considered (for the purpose of this thread only, again debatable, however it was mentioned that this would happen at some stage in the thread) a hidden pair or triple to have at least a quad counterpart..... therefore this counts as a naked triple, but as somebody may see it as a hidden double... it scored a 0 on both.....only one hidden triple came out

tarek

by **gsf** » Mon Apr 17, 2006 10:48 pm

tarek wrote:If you spot the hidden double first.....This is what you see

thanks, that cleared it up
a technique order would handle this case
(e.g., apply hidden double before naked triple)
but it shows that an definitive ranking is not easy to compute

a stepwise approach (1 technique application then restart)
would have row/col/box scanning order bias

a batch approach (all applications for one technique in one step then restart)
would count superfluous applications (maybe it only needed 1 x-wing instead of 10...)

the stepwise approach, even with its bias, seems better, because at least
it attempts to count required technique applications
I'll set up a background test to see how bad the bias is for the superior 100

by **Ruud** » Mon Apr 17, 2006 11:22 pm

gsf wrote:a batch approach (all applications for one technique in one step then restart)
would count superfluous applications (maybe it only needed 1 x-wing instead of 10...)

I have been struggling with this since I introduced a rating function in my solver.

The bias can also introduce superfluous applications. Suppose the solver reports 2 consecutive X-Wings.

When you change the bias from row-col-box to box-col-row and from digit 1-9 to digit 9-1 and RBC index 0-8 to 8-0 (total reversal) then it should require the same number of X-Wings, otherwise the 1st was superfluous.

Of course, this slows the solver down again, but it makes the rating more reliable, unless you have 3 consecutive X-Wings, where the middle one is superfluous...

Ruud.

by **tarek** » Mon Apr 17, 2006 11:46 pm

gsf wrote:a stepwise approach (1 technique application then restart)
would have row/col/box scanning order bias

a batch approach (all applications for one technique in one step then restart)
would count superfluous applications (maybe it only needed 1 x-wing instead of 10...)

What I've done was a batch approach, each BATCH of similar techniques scoring as 1 technique (even if not superflous)

Changing the order of techniques used would insure consistancy of techniques (as in the naked triple, hidden double example)..... if the hidden double had a naked quad counterpart, it would still be there upon changing the order of techniques & therefore would count as A NEEDED hidden double (in the superior sense).

so I FOR SURE could say that puzzle #031 NEEDS 3 x-wings to solve (it may need more) but I can't be wrong on my statement....as you would need 3 batches of xwings to solve when the xwing is the last technique to be used.............

it is easy when it comes to hidden triples & x-wings for the Superior thread....

it will become more difficult & more time consuming on more difficult puzzles with more techniques used....

tarek

by **tarek** » Wed Apr 19, 2006 12:00 pm

this thread has amongst other things provided some good excercises (for both programmers & manual solvers).....

Look at this position in Red Ed's #069,
how many x-wings can you see?
how many eliminations can you achieve?
which ones leave only singles afterwards?

Code: Select all: *-----------------------------------------------------------------* | 45 3 1 | 2 456 7 | 568 9 58 | | 7 6 58 | 58 9 3 | 4 2 1 | | 458 2 9 | 1 4568 456 | 3 56 7 | |---------------------+---------------------+---------------------| | 1 9 258 | 568 3 256 | 7 56 4 | | 258 7 2458 | 9 24568 1 | 2568 3 258 | | 6 45 3 | 458 7 245 | 258 1 9 | |---------------------+---------------------+---------------------| | 3 45 7 | 456 2456 9 | 1 8 256 | | 9 8 6 | 7 1 25 | 25 4 3 | | 25 1 245 | 3 456 8 | 9 7 56 | *-----------------------------------------------------------------*

Next.......

from Red Ed's #045
how many Hidden doubles can you see?
how many eliminations can you achieve?

Code: Select all: *--------------------------------------------------------* | 347 1 34 | 5 47 29 | 69 68 289 | | 79 6 2 | 3 78 89 | 5 1 4 | | 49 5 8 | 1249 246 1269 | 3 7 29 | |------------------+------------------+------------------| | 34 47 3469 | 78 1 38 | 4679 2 5 | | 12 27 139 | 6 5 4 | 179 38 89 | | 5 8 1346 | 27 9 23 | 1467 36 67 | |------------------+------------------+------------------| | 18 3 7 | 1489 468 1689 | 2 5 16 | | 6 9 5 | 12 23 7 | 8 4 13 | | 128 24 14 | 18 36 5 | 67 9 367 | *--------------------------------------------------------*

& Again from the same puzzle....

how many Hidden Tri[ples can you see?
how many eliminations can you achieve?

Code: Select all: *-----------------------------------------------------------------* | 3479 1 3469 | 5 24678 2689 | 69 68 2689 | | 79 679 2 | 3 678 1689 | 5 168 4 | | 49 5 8 | 1249 246 1269 | 3 7 1269 | |---------------------+---------------------+---------------------| | 3479 4679 3469 | 78 1 38 | 4679 2 5 | | 12379 279 139 | 6 5 4 | 179 138 13789 | | 5 8 1346 | 27 9 23 | 1467 1346 1367 | |---------------------+---------------------+---------------------| | 1489 3 7 | 1489 468 1689 | 2 5 16 | | 6 249 5 | 1249 234 7 | 8 134 13 | | 1248 24 14 | 1248 23468 5 | 1467 9 1367 | *-----------------------------------------------------------------*

tarek

by **Red Ed** » Wed Apr 19, 2006 3:29 pm

Personally I find the naked quads much easier to see than their hidden triple counterparts in that puzzle.

The New Sudoku Players' Forum

The SUPERIOR thread