[SciPy-Dev] replacing integrate.cumtrapz with numpy.trapz

[Alan G Isaac]

On Sat, Jul 17, 2010 at 2:42 AM, Ralf Gommers <mailto:ralf.gommers at googlemail.com> wrote:
>  For a small bonus, the numpy version is about 10% faster
>  (tested for several array shapes):
>   >>> a = np.arange(1e4).reshape(500, 20)
>   >>> %timeit np.trapz(a, axis=1)
>  10000 loops, best of 3: 182 us per loop
>   >>> %timeit sp.integrate.cumtrapz(a, axis=1)
>  1000 loops, best of 3: 209 us per loop

That reminds me of a question:
numpy.trapz returns
     add.reduce(d * (y[slice1]+y[slice2])/2.0,axis)
Isn't that equivalent to but slightly less efficient than
    (d * (y[slice1]+y[slice2])).sum(axis=axis)/2.0
?

Alan Isaac