Home Original page

Newbie : innerproduct function from Numarray

David M. Cooke cookedm+news at physics.mcmaster.ca
Mon Jan 26 17:38:22 EST 2004 
At some point, "Ohkyu Yoon" <okyoon at stanford.edu> wrote:

> I have two vectors that have about 2000 elements.
> I need to calculate the innerproducts of those vectors 2000 times where
> one of them cycles left(ie 1234, then 2341, then 3412, etc)
> Basically, I want to calculate A*x, where A is a left-circulant cyclic
> matrix,
> and x is a vector.
>
> I tried it two ways.
> vector1 & vector2 are lists.
                        ^^^^^^

This is your problem: everytime you pass these to innerproduct, it
turns them into arrays (which, with the current version of numarray,
is a relatively expensive operation).

Try something like

import numarray as na
output = []
vector1 = na.array(vector1)
vector2 = na.array(vector2)
temp = na.concatenate((vector1,vector1)) # temp is twice the length of vector1
for i in range(2000):
    output.append(na.innerproduct(temp[i:(i+2000)],vector2)

This uses arrays always.

You might be better off explicity constructing your left-circulant
cyclic matrix A, and doing the full matrix multiply -- for this size
of matrix, the multiply is going to be fast (because it's going to be
done by code written in C, and, depending on how you set it up, using
an optimized BLAS routine).

-- 
|>|\/|<
/--------------------------------------------------------------------------\
|David M. Cooke
|cookedm(at)physics(dot)mcmaster(dot)ca
More information about the Python-list mailing list