[Numpy-discussion] Can this be done more efficiently using numpy?

Didrik Pinte lists@dipole-consulting....
Thu Apr 1 03:21:01 CDT 2010


On Wed, 2010-03-31 at 23:13 -0700, Vishal Rana wrote:
> Hi,
> 
> 
>  A calculation which goes like this...
> 
> 
> n = 5
> a = np.arange(1000)
> b = np.arange(n - 1, 1000)
> 
> 
> l = []
> for i in range(b.size):
>     # Absolute difference of n a elements and nth b element
>     x = np.abs(a[i:i + n] - b[i])
>     
>     # Average of x
>     y = x.sum() / n
>     
>     l.append(y)
> 
> 
> It takes a while if I have like 200K records!
> Is there a more efficient way to do this using numpy?

Like this ? 

import numpy
from numpy.lib.stride_tricks import as_strided
n = 5
a = numpy.arange(1000)
b = numpy.arange(n - 1, 1000)


def sum_1(n, a, b):
    # original method
    l = []
    for i in range(b.size):
        # Absolute difference of n a elements and nth b element
        x = numpy.abs(a[i:i + n] - b[i])
        
        # Average of x
        y = x.sum() / n
        
        l.append(y)
    return l

def sum_numpy(n, a, b):    
    ast = as_strided(a, shape=(1000-n+1, n), strides=(8,8))
    diff = numpy.abs(ast - b[:,None])
    y = diff.sum(axis=1) /n
    return y

test1 = sum_1(n,a,b)
test2 = sum_numpy(n,a,b)

assert(numpy.alltrue(test2 == test1))

-- Didrik

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: This is a digitally signed message part
Url : http://mail.scipy.org/pipermail/numpy-discussion/attachments/20100401/0306d113/attachment.bin 


More information about the NumPy-Discussion mailing list