News for hackers. Stuff that matters.
Re:I No Longer Need StackOverflow by zaptheimpaler on Monday February 17, 2014 @07:17AM (#7251693)
Nope. Thats really the point - if we find the best deterministic algorithm, it means we have found the one with the best deterministic running time. By loosening our criteria of running time to expected running time and allowing randomness, we can sometimes do better than deterministic algorithms.
Re:I No Longer Need StackOverflow by TeMPOraL on Monday February 17, 2014 @08:28AM (#7251959)
It doesn't square well with my intuition - if we understand how allowing randomness improves expected running time, then surely we must be able to create a deterministic algorithm that is at least as good as the randomized one.

Am I missing some critical piece of understanding here?

Re:I No Longer Need StackOverflow by zaptheimpaler on Monday February 17, 2014 @10:54AM (#7252596)
Well keep in mind that the inputs to the algorithm can vary. Randomization can help avoid a bad input from throwing your algorithm off. A simple example of how randomization helps is quicksort:

if we pick the pivot to always be the first element, then any sorted array will cause worst-case running time. But by randomizing the pivot, we can do well (in expectation) for any input.

Re:I No Longer Need StackOverflow by ethbro on Monday February 17, 2014 @10:21AM (#7252418)
I'd expect that it's simply that the additional constraints implied by requiring deterministic behavior precludes certain optimizations that can lead to improved running time.

Iow, if you can never do X, then you cannot write any algorithm that may do X.