On Sat, Mar 06, 2021 at 07:46:18AM +0000, David Mertz wrote:
So this "research" is inherently doomed to fail.... UNLESS, you do the research not by actual raw timings, but rather in the sensible way of profiling the specific number of operations in an abstracted way.
Sorry, are you trying to say that the way to determine better algorithms is not by measuring the actual time they take with real data on real machines, but by counting hypothetical abstract operations? Obviously we can and do use Big Oh analyse guide our investigations, but ultimately nothing beats actual timings. I can't begin to tell you how much time wasted trying to write O(N) algorithms in Python for some task when Python's O(N log N) sort was faster for any size of data I could create on my machine. -- Steve