How to reference a slice/view of a numpy array

Question

An answer I gave earlier raised a question for me: Is it possible to reference a view or slice of a numpy array without repeating a bunch of brackets?

For instance, in the answer I used s=np.argsort(u) and then did all my calculations on a 'virtually' sorted u[s]. I've had situations where I then needed a boolean mask of that array, giving something akin to u[s][mask]. For bigger data I might have a mask of a mask of a mask . . . until things start to look like the end of an episode of Scooby Doo.

But if I assign that array to a variable b=a[s][mask] and change b, a does not change, so I end up carrying a pile of brackets through my calculations. I've tried various arrangements of uv=u.view()[s] but it seems .view() only makes a view of the whole array. Is there another method I am missing?

The issue is [mask]. Since it makes an arbitrary subset of the array a, it always returns a new array, not a view. Can you avoid making a mask in the first place? — Mad Physicist
– Mad Physicist, Commented Dec 14, 2016 at 8:27

John Zwinck · Accepted Answer · 2016-12-14 08:39:12Z

2

You may not be able to solve the simple case of u[s] but in more complex cases like u[s][mask] you can:

t = s[mask]
u[t] # same as u[s][mask]

That is, you can simplify your mask to a single variable, but you may not be able to get rid of it completely, unless perhaps you want to write your own wrapper class with __getitem__ and __setitem__.

answered Dec 14, 2016 at 8:39

John Zwinck

252k44 gold badges346 silver badges459 bronze badges

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

2 Comments

Daniel F Over a year ago

But if I only have masks? I guess I could do t=np.indices(u.shape)[mask1][mask2][mask3]. I assume it ends up being the same problem pandas has where at some layer of abstraction it get tough to tell the difference between a slice and a copy without something like .loc to differentiate

John Zwinck Over a year ago

@DanielForsman: Yeah. In your question you mentioned s=np.argsort(u) which is not a mask, but an array of indices. I worked from that assumption. If you have lots of masks you can compose them and just use the composed version on the original array.

Collectives™ on Stack Overflow

How to reference a slice/view of a numpy array

1 Answer 1

2 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

2 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related