how to vectorize a function whose argument is a vector in numpy

Question

I have an numpy array R of dimension (n, n, n, 3) and a function f which takes a 1-D vector to a scalar. I need a new array A whose relationship with R is

A[i, j, k] = f(R[i, j, k, :])

How can I do this in numpy without three for statements.

This depends on what f is. Hopefully, it'll be as simple as adding axis arguments in a few places. — user2357112
– user2357112, Commented Feb 19, 2016 at 1:06
f is so complicated that I don't even want to post it here. — atbug
– atbug, Commented Feb 19, 2016 at 1:08
Anther 'trick' is to reshape R to 2d, so you index on just one axis, and slice the other. Then reshape the result back. In other words, temporarily 'flatten' the i,j,k dimensions. It won't save iteration steps, but should simplify the code. — hpaulj
– hpaulj, Commented Feb 19, 2016 at 5:56

user2357112 · Accepted Answer · 2016-02-19 01:17:39Z

2

Ideally, you'd do this by changing the implementation of f to use techniques that handle high-dimensional input appropriately. For example, you might change np.sum(whatever) to np.sum(whatever, axis=-1) to get a sum over the last axis instead of the whole array. This would produce the most efficient results, but it might be difficult or impossible, depending on f.

The slower, much easier answer is np.apply_along_axis:

A = np.apply_along_axis(f, -1, R)

This is prettier than 3 for loops, but it probably won't be any more efficient.

answered Feb 19, 2016 at 1:17

user2357112

286k32 gold badges490 silver badges569 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

atbug Over a year ago

Is np.vectorize more efficient than for loops?

hpaulj Over a year ago

Plus vectorize is tricky to apply when the function takes a 1d array instead of a scalar.

Collectives™ on Stack Overflow

how to vectorize a function whose argument is a vector in numpy

1 Answer 1

2 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

2 Comments

Your Answer

Sign up or log in

Post as a guest

Related