Still slower and worse uses 2x the memory for the intermediate temporary
array.
I propose allowing implicit reductions with ufuncs. Specifically if out is
provided with shape[axis] = 1, then pass it on to the ufunc with a stride
of 0. That should allow this to work:
x = np.arange(10)
def add_square_diff(x1, x2, x3):
return x1 + (x2-x3)**2
result =np.zeros(1)
np.frompyfunc(add_square_diff, 3, 1)(result, x, np.mean(x), result)
Essentially it creates a reduce for a function which isn't binary. I think
this would be generally useful. For instance, finding the min and max in
one pass would be nice:
def minmax(x1, x2, x3):
return min(x1,x3), max(x2,x3)
minn = np.array([np.inf])
maxx = np.array([-np.inf])
np.frompyfunc(minmax, 3, 2)(minn, maxx, x, minn, maxx)
Note it also allows for arbitrary initial values or identity to be
specified, possibly determined at run time. I think this would make ufuncs
even more universal.
On Mon, Nov 14, 2016 at 3:38 AM, Jerome Kieffer <jerome.kief...@esrf.fr>
wrote:
> On Fri, 11 Nov 2016 11:25:58 -0500
> Matthew Harrigan <harrigan.matt...@gmail.com> wrote:
>
> > I started a ufunc to compute the sum of square differences here
> > <https://gist.github.com/mattharrigan/6f678b3d6df5efd236fc23bfb59fd3bd>.
> > It is about 4x faster and uses half the memory compared to
> > np.sum(np.square(x-c)).
>
> Hi Matt,
>
> Using *blas* you win already a factor two (maybe more depending on you
> blas implementation):
>
> % python -m timeit -s "import numpy as np;x=np.linspace(0,1,int(1e7))"
> "np.sum(np.square(x-2.))"
> 10 loops, best of 3: 135 msec per loop
>
> % python -m timeit -s "import numpy as np;x=np.linspace(0,1,int(1e7))"
> "y=x-2.;np.dot(y,y)"
> 10 loops, best of 3: 70.2 msec per loop
>
>
> Cheers,
> --
> Jérôme Kieffer
> _______________________________________________
> NumPy-Discussion mailing list
> NumPy-Discussion@scipy.org
> https://mail.scipy.org/mailman/listinfo/numpy-discussion
>
_______________________________________________
NumPy-Discussion mailing list
NumPy-Discussion@scipy.org
https://mail.scipy.org/mailman/listinfo/numpy-discussion
