Improving the timing even more about fast-pandas

[mmngreco]

Yes, actually I forked this project to work on that point here. The reason: if the object passed to a numpy function is not a numpy object then it tries to access method with the same function name that you are trying to use. So when it does run np.sum(df['a']] actually it uses the sum of pandas instead.

You can see that in the numpy code.

Times are fairer with numpy.

PS: Sorry for my english. :-(

[LuckyJosh]

There is no need to apologize, your english is fine. 😉

[mm-mansour]

Actually np.mean(df["A"]) is slower because it calls df["A"].mean(), just as @mmngreco mentioned.
I have added df["A"].values to mean and all other functions, check them out and thanks a lot for the suggestions.