The hardest sudokus

Everything about Sudoku that doesn't fit in one of the other sections

Postby champagne » Sat Sep 13, 2008 11:53 pm

coloin wrote:
When champagne gets his solver on these what brand of memorable labeling can we use for the likes of colx467 !!

C


I just open my computer after one week off, I see that after three months lethargy, fresh stuff is coming.

I hope I'll have time to-day to start sorting of these lots and good news:D

champagne
champagne
2017 Supporter
 
Posts: 6285
Joined: 02 August 2007
Location: France Brittany

Postby champagne » Sun Sep 14, 2008 12:15 am

coloin wrote:Here are new puzzles that have not been demonstrated [I believe] before
Top 50 in each rating


C


As you know, I process files with puzzles starting in position 1.

If you can send me your lists in line with that specification, it woill help.

If not, I will have to reshape them

champagne
EDIT
ps: same is valid for tarek contribution

EDIT2 I could manage it easily, first lot is runnng.
champagne
2017 Supporter
 
Posts: 6285
Joined: 02 August 2007
Location: France Brittany

Postby champagne » Sun Sep 14, 2008 2:55 am

Before giving back results for the last lots of "hardest", some comments on the status of the solver.

Normally,I have not changed recently the process for hardest puzzles. I am working oneasier puzzles to find shorter paths.
Nevertheless, I changed some poor code in common areas.
The processing time has decrased and in my top list one puzzle has been downgraded (col809 herebelow).
In fact, all puzzles having unitary steps where over valued due to that poor coding.
I was reluctant to put col809 at the level of silverplate and Golden nugget due to the relatively small printsize. I feel confortable with the new figures.
I summarize the comparative reference four puzzles

Code: Select all
                ms   sec
             count  time printsize
Silver Plate   203  87.2 669K
Golden Nugget  125  63.9 567K
col809#         62  31.6 362K
Easter Monster 156   6.7 132K


EM remains about ten times below Silver Plate and Golden Nugget as far as processing time is concerned and about 1/5 in print size.

The three colums are giving the classical set of rough figures runnng the program under windows in multi tasking context:

c1: counting the number of solutions (milliseconds)
c2: solving the puzzle (seconds)
c3: raw printsize in Kbytes

Now the results for the first lot proposed by coloin.
Average processing time is 17.2s far above EM, and the hardest puzzles are not so far from SP and GN.
I give the full list mainly because you will see several times sequences with nearly the same figures.

Code: Select all
               ms   sec         
            count  time printsize
 #   colx001  141  14.9  280K
 #   colx002   78  17.8  329K
 #   colx003   47  13.1  255K
 #   colx004   94   9.5  219K
 #   colx005   78   9.5  219K
 #   colx006   94   7.7  176K
 #   colx007  235   7.7  129K
 #   colx008  172  21.3  394K
 #   colx009  110  13.9  240K
 #   colx010   78  15.7  306K
 #   colx011   78  15.8  306K
 #   colx012   63   6.7  151K
 #   colx013   47  10.2  203K
 #   colx014   62  14.0  292K
 #   colx015  125  23.2  344K
 #   colx016   78  17.8  330K
 #   colx017   78  17.9  330K
 #   colx018   47  17.5  213K
 #   colx019   62  25.3  361K
 #   colx020  109  29.6  415K
 #   colx021   78  18.7  304K
 #   colx022  125  15.7  265K
 #   colx023   47  14.6  261K
 #   colx024  110  20.1  371K
 #   colx025  109  20.4  373K
 #   colx026   93  16.1  327K
 #   colx027   78  16.1  327K
 #   colx028   47  13.9  230K
 #   colx029   93  11.4  223K
 #   colx030  140  16.7  243K
 #   colx031   46  11.2  217K
 #   colx032   62  14.5  210K
 #   colx033   78   9.9  186K
 #   colx034  140  20.7  263K
 #   colx035   78  19.0  288K
 #   colx036   62   6.4  181K
 #   colx037   47  11.1  247K
 #   colx038  140  27.2  329K
 #   colx039   62  24.8  374K
 #   colx040   78  18.1  306K
 #   colx041   94  17.0  313K
 #   colx042   78  17.0  313K
 #   colx043   78  29.2  347K
 #   colx044  234  35.4  458K
 #   colx045  203   8.3  124K
 #   colx046   78  19.4  321K
 #   colx047   46  14.6  260K
 #   colx048   78  45.6  483K
 #   colx049  109   6.8  138K
 #   colx050   94  21.5  278K




EDIT
Lot 2 of coloin seems slightly easier that the first one. I just picked up the best (cut off 20 seconds). Three puzzles were in the first lot.

Code: Select all
#   colx134  203  37.3  339K
#   colx1274  94  22.2  322K
#   colx039   62  24.8  374K
#   colx131  125  27.1  319K
#   colx102  125  34.2  380K
#   colx180   94  34.6  321K
#   colx015  125  23.2  344K
#   colx247  109  25.7  325K
#   colx094  110  20.0  275K
#   colx1922 250  30.4  453K
#   colx1107  47  20.1  299K
#   colx1173  63  26.4  363K
#   colx044  234  35.3  458K
#   colx1890 172  37.4  417K


The third lot does not bring any new interesting puzzle



champagne
champagne
2017 Supporter
 
Posts: 6285
Joined: 02 August 2007
Location: France Brittany

Postby champagne » Sun Sep 14, 2008 5:59 am

Here the result for the first 50 puzzles of tarek.

A key point: two puzzles have not been solved: number 1 and number 6.

After a quick check, it could be that they can not be solved with the set of rules used at level 4 of my process.
Number 1 is nearly immediatly locked, number 6 is locked after a small number of steps.
The rest is in average easier than coloin's lot, but number 8 is very close to the two stars. Number 10 is not that bad.
It would be interesting to know whether Alan Barker can solve numbers 1 and 6.

Code: Select all
#tarx0001 391   
#tarx0002 422   27.5  306K
#tarx0004 109    6.3  141K
#tarx0003 125    8.2  170K
#tarx0005 109    9.3  186K
#tarx0006 235 
#tarx0007 157    7.2  169K
#tarx0008 281   49.0  459K
#tarx0009 171   12.2  184K
#tarx0010 172   32.7  420K
#tarx0011 359    7.3  165K
#tarx0012  94    4.9  87K
#tarx0013 219   16.8  292K
#tarx0014  78    6.8  168K
#tarx0015  47   17.7  291K
#tarx0016  32   13.5  184K
#tarx0017  94   12.1  264K
#tarx0018  63    6.1  132K
#tarx0019  93    9.5  212K
#tarx0020 125    5.6  119K
#tarx0021 140   11.7  203K
#tarx0022 125   15.9  249K
#tarx0023 187   29.5  356K
#tarx0024 156    7.5  158K
#tarx0025  78   26.4  271K
#tarx0028  78    3.1  61K
#tarx0026  47    7.5  140K
#tarx0027 203    8.6  181K
#tarx0030  94    6.8  174K
#tarx0031  62    6.1  149K
#tarx0029 125    7.0  157K
#tarx0032 110   28.0  299K
#tarx0033 110    7.8  168K
#tarx0034 156    6.1  124K
#tarx0035 188   24.5  224K
#tarx0036  62   14.3  235K
#tarx0037  63    4.2  103K
#tarx0038 125    5.4  134K
#tarx0039  78   11.3  231K
#tarx0040 235   11.7  201K
#tarx0041 140    7.0  103K
#tarx0042  93   25.0  380K
#tarx0043  78   11.6  234K
#tarx0044 234    6.9  156K
#tarx0045 156    5.9  115K
#tarx0046  46    6.1  129K
#tarx0047  15    8.5  148K
#tarx0048  46    6.8  155K
#tarx0049  78   13.7  240K
#tarx0050 109    6.5  146K
#tarx0050 109    6.5  146K
#tarx0050 109    6.5  146K


end of tarek list to come

champagne
champagne
2017 Supporter
 
Posts: 6285
Joined: 02 August 2007
Location: France Brittany

Postby champagne » Sun Sep 14, 2008 6:47 am

The remaining part of tarek's list.
Average processing time 11.5 second for 109 puzzles.

Good surprise, a new challenger to beat SP and NG (number 119).
In my ranking, it stays in between with a processing time higher than GN, but a shorter print. To be investigated later.
No new puzzle defeating the solver.
Here after the short list with the same cut off (20 seconds) as for coloin sublists 2 and 3.

Code: Select all
#tarx0053  140   30.5 print 388K
#tarx0054  157   29.2  335K
#tarx0069  125   32.5  393K
#tarx0074   94   20.2  190K
#tarx0075  141   28.5  310K
#tarx0091  219   26.0  280K
#tarx0104  110   33.9  424K
#tarx0105  109   30.0  392K
#tarx0108  141   26.9  276K
#tarx0110  188   24.7  321K
#tarx0119  297   71.4  539K
#tarx0121  140   27.9  353K
#tarx0125   94   26.7  342K
#tarx0135  297   46.7  401K
#tarx1199   78   22.6  276K


Unless I missed something, I have covered the request.

champagne
champagne
2017 Supporter
 
Posts: 6285
Joined: 02 August 2007
Location: France Brittany

Postby coloin » Sun Sep 14, 2008 2:29 pm

Thanks - more than covered !

I will leave it up to tarek to come up with a "memorable" name tag for any of his puzzles which merit it.

I had continued along the lines of precious metals and an accompaning word. The two initials need to be unique of course.

However I did like the "cloudy bay" - shame the puzzle didnt live up to the vintage !

C
coloin
 
Posts: 1715
Joined: 05 May 2005

Postby tarek » Sun Sep 14, 2008 3:34 pm

Thanks champagne for your review ....

I'm not sure if tarx0001 & tarx0006 being unsolvable according to your solver doeas mean they are above the rest ....

I would suggest that a review of the solver first. It often turns out to be a bug.

If it was/wasn't due to the solver & your solver in the future was updated so that they are solvable then a re-run through EVERYTHING is needed.

I think that good puzzles are worth naming. so will look up some names:D

I discovered an isomorph in my latest lists:
Code: Select all
.....6..1.2.....7...6.5.8.....49......45..9..3..8.......86..5...1.....2.7.......3#tarx0125
is isomorphic to
.......39.....1..6..3.9...5..5.6...81....2....7.4.......6.8..5..2.7.....4.....8..#colx1107
I will delete that entry from the list.

I have enjoyed this search, hopefully more would jointhe search .... As the tools are there & there is plenty for eveybody.

tarek
User avatar
tarek
 
Posts: 2650
Joined: 05 January 2006

Postby champagne » Sun Sep 14, 2008 11:52 pm

tarek wrote:Thanks champagne for your review ....

I'm not sure if tarx0001 & tarx0006 being unsolvable according to your solver doeas mean they are above the rest ....

I would suggest that a review of the solver first. It often turns out to be a bug.



Fully agreed.

Nevertheless, mostoften, a bug turns to a wrong clearing, easy to see. Tougher to detect is silence.

Here it seems that (at least for tarx0001) there is nearly no stuff in the families recognized by my solver (bi balues including groups; AHS/AC;AC2). But I will study both and give a better feeling later.

Anyway, we shoul receive other contributions from all guys having tried to face hardest puzzles. It will be helpfull.

If these puzzles are requesting additional stuff to be solved, it will be tougher to handle, unless a brillant idea comes. (I have small hopes out of improvements I am implementing, but I doubt it will be enough to crack theese two)

champagne
champagne
2017 Supporter
 
Posts: 6285
Joined: 02 August 2007
Location: France Brittany

Postby champagne » Tue Sep 16, 2008 10:16 am

tarek wrote:If it was/wasn't due to the solver & your solver in the future was updated so that they are solvable then a re-run through EVERYTHING is needed.tarek


To show again the difficulty of rating for hardest puzzles, I can give my last experience (dated yesterday).

Updating of the solver in my case is a permanent situation. I do not stay with releases to have a stable result. Rating is not my target.

As a matter of fact, each time I make a change in the solver, I have to re run my sample file, and, in case it is a key move for hardest puzzles, to re run my own file of hardest.

Checkng the results for "Fata Morgana" and "trompe l'oeil", I noticed a small anomaly that could have turned to a bug.

I corrected the program and I revised the sequence around that anomaly in a kind of fine tuning in harmony with what I have done recently on easier puzzles.

two points:

1) Results are relatively sensitive to that fine tuning,
2) I my "final" (temporary) version, processing time and printsize decreased significantly in average with a very significant exception in my sample file : " AI ETANA "

and for sure, the relative ranking will change slightly.

We can imagine in a final state to have options to adjust the fine tuning puzzle by puzzle (my view is that a puzzle should be rated for the "best solution" in that field of hardest puzzle).

For the time being, it is not a priority for me, so I can not do more than to deliver results non stable ove time..


Champagne
champagne
2017 Supporter
 
Posts: 6285
Joined: 02 August 2007
Location: France Brittany

Postby tarek » Tue Sep 16, 2008 2:52 pm

champagne wrote:Checkng the results for "Fata Morgana" and "trompe l'oeil", I noticed a small anomaly that could have turned to a bug.
I'm glad that some good use came out of these puzzles. Our thirst to know the hardest puzzle is probably the cause behind demanding so much from you:D .

When I programmed my solver (Late 2005 - Early 2006) the top1465 were the obstacle to cross. I can't tell you how many times I ran through that list each time I updated the solver because I lost count

As with SE, each time a version changes, the ratings change ... therefore the correct version needs to be next to the rating program cited.

tarek
User avatar
tarek
 
Posts: 2650
Joined: 05 January 2006

Postby tarek » Sat Sep 20, 2008 5:11 am

I guess somebody has to step in at some stage,

Do you think that we should start a new thread about this subject with the head post updated by an active member .....

I hjave been reluctant to start it because of the responsibilty attached to it (I'm still behind with the Ultimate fish guide)... but I suppose ravel hasn't been here since his 999th post & the head post is REALLY outdated now ....

The head post will have the top 10 from each rating program (RMS will have to be dropped out due to absence of active participation by the only person who can perform that rating)....

Any suggestions about the rating programs that should be included ?

my 1st suggestion would be the freely available rating programs .... q1, q2, sx9, sxt & SE121

The basis of the list would be the q2 taxonomy list & any more recent posts from coloin & me.

if all is happy with that I can perform a sxt (1000 2) rating of the q2 taxonomy & leech the q2 q1 in it to start with ........

The problem would then be to know what SE121 ratings are .... We have to go with what is known.

A good idea would be for the users to rate their own puzzles (I know that can take ages) , but it could be the price needed for inclusion in the lists:!:

tarek
User avatar
tarek
 
Posts: 2650
Joined: 05 January 2006

Postby JPF » Sat Sep 20, 2008 5:35 am

I had the same idea...:)

  • start a new thread
  • update the top 20(?) puzzles for each public rating q1, q2, sx9, sxt, ER.
  • Each submitter gives the ratings
So we need somebody to :
  • open the thread
  • check the isomorphs
  • check the ratings
  • update the lists at the top of the thread
It would be nice if the beginning of the thread would remind the way to get the different public ratings.

What about our best friend g.r.emlin ?

JPF
JPF
2017 Supporter
 
Posts: 3754
Joined: 06 December 2005
Location: Paris, France

Postby ttt » Sat Sep 20, 2008 8:39 am

JPF wrote:I had the same idea...:)

  • start a new thread
  • update the top 20(?) puzzles for each public rating q1, q2, sx9, sxt, ER.
  • Each submitter gives the ratings
So we need somebody to :
  • open the thread
  • check the isomorphs
  • check the ratings
  • update the lists at the top of the thread
It would be nice if the beginning of the thread would remind the way to get the different public ratings.

JPF


For me, that is seemed to right...!
Last private message I received from ravel on Sun Nov 11, 2007 5:14 am. I think that when comes 99th... I should follow him...

ttt
ttt
 
Posts: 185
Joined: 20 October 2006
Location: vietnam

Postby gsf » Sat Sep 20, 2008 10:05 pm

JPF wrote:I had the same idea...:)

  • start a new thread
  • update the top 20(?) puzzles for each public rating q1, q2, sx9, sxt, ER.
  • Each submitter gives the ratings
So we need somebody to :
  • open the thread
  • check the isomorphs
  • check the ratings
  • update the lists at the top of the thread
It would be nice if the beginning of the thread would remind the way to get the different public ratings.

What about our best friend g.r.emlin ?

g.r.emlin just learned how to re-edit posts
submissions could be collated and edited into one post
would submissions be to the thread or via pm?
gsf
2014 Supporter
 
Posts: 7306
Joined: 21 September 2005
Location: NJ USA

Postby StrmCkr » Sat Sep 20, 2008 10:49 pm

they should be to a locked thread.

edited and updated via Pm:)
Some do, some teach, the rest look it up.
User avatar
StrmCkr
 
Posts: 784
Joined: 05 September 2006

PreviousNext

Return to General