RW wrote:gsf wrote:one rating for all is the ultimate goal
That's right. I'd prefer one that spreads out the harder puzzles quite a bit, squeezing all puzzles that need propositions into the top 1% of the scale is not right. The scale doesn't really reflect the big differences in difficulty that appear near the top end. As I said there's a big range between 30000-99000 that rarely is found (only 30/1000 randomly generated puzzles belonged to this category). A large chunk of this range should belong to puzzle that can be solved with relatively few and short propositions.
I hedged my bets early on by making the 90000-99999 range exponential
when the dust settles the hardest should be >= 90000 with a bigger spread like you suggest
for now I haven't done micro-recalibrartions because the rating is quoted in the forums
RW wrote:But hey, now I feel bad for all the time pointing out (the very few and minor) flaws I can find in your program. It's a superb program and we all thank you for providing us with such a great tool. Don't forget that!
thanks
if you look at the --man docs for -R and -qhardest -q1 -q2
you'll see that they're just shorthand for long/complex option combinations
in particular -q for constraint order and -R expressions for the ratings
the upper case vars in -R correspond to stats collected by the constraint methods applied
the %Q format lists the rating (via the -R expression), the constraint methods used,
and the cumulative stats for the constraint methods used to solve
for example
- Code: Select all
sudoku -q1 -f%Q 5.....3...2.1...7...8.....9.4...7......821......6...1.3.....8...6...4.2...9.....5
prints
- Code: Select all
101 FNP C21.m/S2.p/F8653.17653/N12256.31819/P2.20.13246.9.12.33341.352.265.2/M2.8.25420
where:
C21.m: 21 clues minimal
S2.p: symmetry order 2 "pi" (90 rotation)
P2.20.13246...: P(P0).(P1).(P2)...: look for Pi in the -R rating
M2.8: singles backdoor size 2, 8 backdoors
the <method-id><index> are defined in the --man doc
the ultimate rating (at least for what my solver is capable of producing) is hiding in the stats
(a few more stats may be needed, but the basic -q -R model will remain)
so anyone can play with -q and -R to get it to match the current reality
if you do, set -BX to batch constraint moves and ignore backdoors
