On 3/23/12 5:42 PM, Manfred Nowak wrote:
Andrei Alexandrescu wrote:
In the limit, taking the minimum over infinitely many
measurements of X would yield T.
True, if the thoretical variance of the distribution of T is close to
zero. But horrible wrong, if T depends on an algorithm that is fast
only under amortized analysis, because the worst case scenario will be
hidden.
Wait, doesn't a benchmark always measure an algorithm with the same
input? For collecting a chart of various inputs, there should be various
benchmarks.
Andrei
