[SciPy-user] Vectorization question

[Anand Patil]

>Anand Patil wrote:

>>/Hi all,
/>>/
/>>/I want to make array A from array B like so:
/>>/
/>>/A[t, j, k] = \sum_i B[t, j, i] B[t, i, k]
/>>/
/>>/That is, for each t
/>>/
/>>/A[t,] = dot(B[t,], B[t,])
/>>/
/>>/There's no loopless way to do this in numpy, right?
/>>/  
/>>/
/>You should be able to do this just using
>
>A = dot(B,B)
>
>Because the dot function returns the sum of products over the last 
>dimension of the first argument and the second-to-last dimension of the 
>second argument.
>
>-Travis


The output array I'm looking for is rank-3, but as I understand them dot and tensordot can only ever return even-rank arrays:

In [1]: from numpy import zeros, dot

In [2]: B=zeros((4,3,3))

In [3]: dot(B,B).shape
Out[3]: (4, 3, 4, 3)

In [4]: 

Thanks,
Anand