Oops. I accidentally sent this just to Simon; it was supposed to go
to the list:
Simon Kröger [email protected] writes:
Well, what’s an iteration in your code?
I found a solution (the same) to your sentence above after
255062 full circles over all 26 letters. (changing only some
of them each time) It took 36s on my laptop. (Pentium M 2.1 GHz)
I ran it again and it took 134s and did 10105853 changes to
the original guess (counting each changed count separately).
I think here we see part of the reason why Simon’s code works faster.
He’s using a different search algorithm than we are. What I’m doing,
and I suspect that what Christoffer is doing as well, is taking an
initial set of 26 letter counts, figuring out the sentence that works
out to,(*) and then with the 26 counts in the worked out sentence
choosing a random value in between to get a new slate of letter
However, it would appear that Simon is recalculating the whole
sentence after each letter re-adjustment, rather than recalculating it
only after adjusting all 26 counts. This apparently leads to a much
Obviously (at least with my algorithm) it depends on luck.
And for the total Time it takes it depends on your PC and
My PC specs out at faster than yours, even if I am running ruby under
cygwin, that doesn’t turn sub-minute times into over an hour. I think
the “adjust one and recalculate” approach is just a faster approach.
(*) Or something equivalent after optimization. I don’t actually
build the sentence each time. But the point is that I do adjust the
frequencies of all letters between recalculating the letter counts
found in the sentence.
Having now changed to that method, (change one letter and recalculate)
I’m now getting sub-minute times reasonably frequently. I’m still not
getting anywhere near the times quoted, but I’m going to try some
other optimizations to see what I can get.