copying a list into part of a numpy array

Question

I have a list that varies in size and I would like to copy it over a numpy array. I already have a way to do but I would like to see if there is a more elegant way.

Let's say I have the following:

import numpy as np
data = np.full((10,), np.nan)

# I want to copy my list into data starting at index 2
def apply_list(my_list):
    data[2:2+len(mylist)] = mylist

# Example
tmp = [1, 2, 3, 4]        # can vary from 2 to 8 elements
apply_list(tmp)

After this, I expect data to look like this:

[nan, nan, 1, 2, 3, 4, nan, nan, nan, nan]

Please keep in mind that len(mylist) can range from 2 to 8.

I am marking unused places with NaN and data has been preallocated and should always be size=10, regardless of the size of my_list. For that reason, simply appending will not work.

I particularly don't like much doing 2:2+len(mylist). Is there a nicer/cleaner way of doing this?

Is there some way of combining this with some iterator/generator? (obviously without having to iterate in plain python). I know about np.fromiter but not sure how to connect that to the assignment — Juan Leni
– Juan Leni, Commented May 12, 2017 at 18:16
well.. both :). I am learning more about numpy and I came up with this problem. If I can improve style, it would be great. But clearly I dont want to have a performance hit. — Juan Leni
– Juan Leni, Commented May 12, 2017 at 18:18
You case matches the 2nd example in this pararagraph: docs.scipy.org/doc/numpy/user/…, x[2:7] = np.arange(5). Assignment to a list works the same way - except that the RHS slot does not have match in size. Array size is fixed, list size is not. — hpaulj
– hpaulj, Commented May 12, 2017 at 18:22

MSeifert · Accepted Answer · 2017-05-12 18:16:15Z

2

I'm not aware of any numpy-function that could simplify this. However you could wrap it as function so the complexity is hidden:

def put(arr, subarr, startidx):
    arr[startidx:startidx+len(subarr)] = subarr
    return arr

or with sequential indexing (not recommended):

def put(arr, subarr, startidx):
    arr[startidx:][:len(subarr)] = subarr
    return arr

You could also pad your mylist with NaNs:

np.pad(np.array(mylist, dtype=float), 
       (2, 8-len(mylist)), 
       mode='constant', 
       constant_values=np.nan)

edited May 12, 2017 at 18:16

answered May 12, 2017 at 18:08

MSeifert

154k41 gold badges356 silver badges377 bronze badges

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

3 Comments

Juan Leni Over a year ago

sorry to ask about the "non recommended" one.. but it looks interesting. I never thought about it.. Why do not say is not recommended?

MSeifert Over a year ago

I think padding isn't really as performant, the other ones will be faster. I meant I wouldn't recommend the second approach becauseit just avoids the startidx+ by adding a not-so-readable double slicing operation.

hpaulj Over a year ago

np.pad is a complex function that does a lot of concatenates. And at the compiled code level, concatenate does the same sort of allocate and assign operation that the OP is trying to avoid.

Collectives™ on Stack Overflow

copying a list into part of a numpy array

1 Answer 1

3 Comments

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

3 Comments

Related