Good answer.

Mozzila documentation state this: Deprecated: This feature is no longer recommended.

I have a gradient descent kata for you http://www.codewars.com/kata/linear-regression-with-one-variable/ruby :-)

It's in Ruby - let me know if you'd like a js version

uhm, Octave if I'm not mistaken. it had some usefull built in stuff for Matrices and such, but the language itself was just weird, it wasn't bad, it just felt... wrong (I can't remember why, probably stuff like 1 based arrays and stuff like that)

+1 would love a kata like that as well. Just out of curiosity, in which language was it in your CS course ?

I would include this as well:

Test.assertEquals(largestDifference([9,4,1,2,3,0,-1,-2]), 2);

As an aside the test errors are clearer when using assertEquals.

You could also look into converting the number into real binary first :) and trying to find a pattern on converting that to 'negabinary'.

that said, most solutions haven't done it that way, but it is a possible route.

Added

To help improve on this example, it seems that the following test would also pass:

Test.expect (largestDifference([9,4,1,10,3,4,0,-1,-2]) == 4); // data[1] <= data[5), so 5 - 1 = 4
Test.expect (largestDifference([3,2,1]) == 0); // never finds data[n<i] <= data[i]

I believe "some" breaks on the first hit, where as "every" has to go over every element in the array, making it a lot slower for higher numbers. (not sure if this is the case with javascript, but it would make sense)