Padding large numpy arrays

Question

Another numpy array handling question: I have an approx. 2000³ entries numpy array with fixed size (that I know), containing integers. I want to pad the array with another integer, so that it's surrounded in all dimensions. This integer is fixed for the whole padding process.

example (2D)
1----->000
       010
       000

I have two ideas, leading to that result:

Creating a larger numpy array, containing the padded values and "slicing" the old area in the padded:
```
padded=np.zeros((z+2,x+2,y+2))
padded[1:z+1,1:x+1,1:y+1]=olddata
```

Using np.insert or hstack,vstack,dstack to add the values:

padded=np.insert(data,0,0,axis=0)
padded=np.insert(data,x+1,0,axis=0) etc.

The Problem is, that all these methods are not in-place and allocate a new array (1.) or copy the old one (2.). Is there a way to do the padding in-place? I know, that since numpy 1.7. there's the numpy.pad module. But that also seems to use some kind of allocation and overriding (like in my 1. way).

@JanneKarila : I would not know how to start with a padded array? I have the rawdata and on some point I have to pad it or don't I? — Dschoni
– Dschoni, Commented Apr 10, 2013 at 11:17
How do you get the raw data into a NumPy array? Perhaps you could read it into a slice inside the padding at that point. — Janne Karila
– Janne Karila, Commented Apr 10, 2013 at 11:22
@JanneKarila I could do that, but I need to create the Padding as an array before, which takes a lot of time and memory (np.ones(2000,2000,2000) for example takes nearly 4 seconds) — Dschoni
– Dschoni, Commented Apr 10, 2013 at 11:36
Instead of np.ones you could use np.empty and slice assignments to set the padding. — Janne Karila
– Janne Karila, Commented Apr 10, 2013 at 11:40

Janne Karila · Accepted Answer · 2013-04-10 12:31:40Z

1

You can't add padding in-place because there is no room for it in the memory layout. You can go the other way though: allocate the padded array first, and use a view into it when accessing unpadded data.

padded = np.empty((2002,2002,2002))
padded[0] = 0
padded[-1] = 0
padded[:,0] = 0
padded[:,-1] = 0
padded[:,:,0] = 0
padded[:,:,-1] = 0

unpadded = padded[1:-1, 1:-1, 1:-1]

answered Apr 10, 2013 at 12:31

Janne Karila

25.3k6 gold badges60 silver badges97 bronze badges

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

Collectives™ on Stack Overflow

Padding large numpy arrays

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related