Characterizing the hardest sudokus

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

Postby ravel » Mon Oct 02, 2006 4:09 pm

Thanks for the explanations, so i can experiment with the options.

For my personal taste, the use of pairs/triples in elimination chains is overrated both in SE and your solver. E.g. an ALS (which eventually could solve a puzzle in a rather elegant way) would not even be detected by SE, if it contains an almost triple, and i suppose it would be rated high by your solver with standard options.
ravel
 
Posts: 998
Joined: 21 February 2006

Postby gsf » Mon Oct 02, 2006 4:32 pm

ravel wrote:Thanks for the explanations, so i can experiment with the options.

For my personal taste, the use of pairs/triples in elimination chains is overrated both in SE and your solver. E.g. an ALS (which eventually could solve a puzzle in a rather elegant way) would not even be detected by SE, if it contains an almost triple, and i suppose it would be rated high by your solver with standard options.

right, adding ALS constraints will change what puzzles are considered hardest

I was approaching the analysis from the other end using the proposition
technique to model solver guessing when the basic constraints run out

there may be a relationship between the proposition constraints required to
solve a puzzle and any non-proposition constraints that also solve the puzzle
gsf
2014 Supporter
 
Posts: 7306
Joined: 21 September 2005
Location: NJ USA

Postby tarek » Mon Oct 02, 2006 5:12 pm

Are the parameters the same for the windows executable version of gsf's solver?

I've been trying to use dos compatible version of what ravel has been using with no success.......


I used suexrat9 last time for screening............. hopefully gsf's corresponds better with your BF method ravel.......


tarek
User avatar
tarek
 
Posts: 2607
Joined: 05 January 2006

Postby ravel » Mon Oct 02, 2006 5:22 pm

tarek wrote:... hopefully gsf's corresponds better with your BF method ravel...
With standard settings i fear not - from what i have tried so far.
ravel
 
Posts: 998
Joined: 21 February 2006

Postby ravel » Mon Oct 02, 2006 6:40 pm

tarek wrote:Are the parameters the same for the windows executable version of gsf's solver?

I've been trying to use dos compatible version of what ravel has been using with no success.......
Being on a windows pc now, i saw that this works:
I wrote this to a file rate.bat:
sudoku -B -f%%f:%%j%%16@%%(V)x%%,%%2(P)x%%,%%5r -q hardest puzzles.dat > ratings.dat

Calling "rate" in a cmd shell the puzzles in puzzles.dat were rated to the file ratings.dat using this format:
Code: Select all
puzzles:1       6 10 99706
puzzles:2       6  5 99660
ravel
 
Posts: 998
Joined: 21 February 2006

Postby tarek » Mon Oct 02, 2006 9:37 pm

Thanx ravel,

It works now....... so now I could have the suexrat9 & gsf's rating as screening methods with SE as a backup.......

Now, could you formulate a screening formula for us to use, so that I would feel less guilty when I post >10 puzzles next time in the hardest sudokus thread:D

This could work to characterise the hardest sudokus also, something like......

Code: Select all
suexrat9>=450 OR gsfr>=99400 OR SE>=9.6


tarek
User avatar
tarek
 
Posts: 2607
Joined: 05 January 2006

Postby ravel » Mon Oct 02, 2006 10:05 pm

tarek wrote:Now, could you formulate a screening formula for us to use, so that I would feel less guilty when I post >10 puzzles next time in the hardest sudokus thread:D
To be honest, i dont know. I know some drawbacks of my rating and some of others. The point is: gsf's rating takes you some 10th of seconds for the hardest, ER some minutes and mine some hours. They all look at different properties, and also manual solvers do. So at the moment i dont know a better way than post the hardest according to suexrat and gsf and see, what Eplainer and my program say to them.
ravel
 
Posts: 998
Joined: 21 February 2006

Previous

Return to General