On 12-Oct-09, at 11:53 PM, per freem wrote:

what i want to do is pick the columns l of p, and multiply each one by the numbers in s. the first step, picking columns l of p, is simply:

cols = p[arange(3), l]

This isn't picking columns of p, this is picking the times at (0, 0), (1, 0), and (2, 1). Is this what you meant? In [36]: p[arange(3), [0,0,1]] Out[36]: array([ 0.2, 0.5, 0.7]) In [37]: p[:, [0,0,1]] Out[37]: array([[ 0.2, 0.2, 0.8], [ 0.5, 0.5, 0.5], [ 0.3, 0.3, 0.7]]) In [38]: p[arange(3), [0,0,1]] * s.reshape(3,1) Out[38]: array([[ 2., 5., 7.], [ 4., 10., 14.], [ 6., 15., 21.]]) In [41]: p[:, [0,0,1]] * s.reshape(3,1) Out[41]: array([[ 2., 2., 8.], [ 10., 10., 10.], [ 9., 9., 21.]]) Notice the difference.

then i want to multiply each one by the numbers in s, and i do it like this:

cols * s.reshape(3,1)

this seems to work, but i am concerned that it might be inefficient. is there a cleaner way of doing this? is 'arange' operation necessary to reference all the 'l' columns of p?

That's about as efficient as it gets, I think.

also, is the reshape operation expensive?

No. It will return a view, rather than make a copy. You could also do cols * s[:, np.newaxis], equivalently. David