Converting list to numpy array

Question

I have managed to load images in a folder using the command line sklearn: load_sample_images()

I would now like to convert it to a numpy.ndarray format with float32 datatype

I was able to convert it to np.ndarray using : np.array(X), however np.array(X, dtype=np.float32) and np.asarray(X).astype('float32') give me the error:

ValueError: setting an array element with a sequence.

Is there a way to work around this?

from sklearn_theano.datasets import load_sample_images
import numpy as np  

kinect_images = load_sample_images()
X = kinect_images.images

X_new = np.array(X)  # works
X_new = np.array(X[1], dtype=np.float32)  # works

X_new = np.array(X, dtype=np.float32)  # does not work

You have a list of lists, where not all lists have the same amount of entries. — Daniel, Nov 10 '14 at 18:28
ok. What is the solution? New to python so please bear with me — Priya Narayanan, Nov 10 '14 at 18:59
Check the type after your `np.array(x)` is successful. If it is float64 which is the default - then you are simply trying to assign a type which is not suitable for your list (or list of lists). — ha9u63a7, Nov 10 '14 at 19:00
possible duplicate of [ValueError: setting an array element with a sequence](http://stackoverflow.com/questions/4674473/valueerror-setting-an-array-element-with-a-sequence) — ha9u63a7, Nov 10 '14 at 19:01
The dtype of np.array(x) uint8. I think it is a list of lists — Priya Narayanan, Nov 10 '14 at 19:02
if it's really uint8, then there should be no problem to convert it into float32. But I doubt this. Can you show more code? — Daniel, Nov 10 '14 at 19:14
it seems unlikely that ``np.array(X).dtype == np.uint8`` but ``np.array(X, dtype=np.float32)`` fails. To check if all lists that you want to put in the array have the same length, you can do ``np.unique([len(x) for x in X])`` — Andreas Mueller, Nov 10 '14 at 20:24

score 141 · Answer 1 · answered Oct 24 '17 at 15:03

141

If you have a list of lists, you only needed to use ...

import numpy as np
...
npa = np.asarray(someListOfLists, dtype=np.float32)

per this LINK in the scipy / numpy documentation. You just needed to define dtype inside the call to asarray.

answered Oct 24 '17 at 15:03

Thom Ives

3,642
3
30
29

score 1 · Answer 2 · answered Feb 23 '23 at 23:16

If you're converting a list into an array, you'll need to make a new copy anyway, so

arr = np.array(my_list, dtype='float32')

also works.

One use case is when my_list is not actually a list but some other list-like object; in which case explicitly casting to list beforehand might be helpful.

arr = np.array(list(my_list), dtype='float32')

For example,

my_list = pd.Series([[1], [2], [3]]).values

np.array(my_list)             # jagged array; not OK
np.array(list(my_list))       # OK
np.array(my_list.tolist())    # OK


my_list = {'a': [1], 'b': [2]}.values()
np.array(my_list)             # jagged array; not OK
np.array(list(my_list))       # OK

To cast nested list into an array, the shapes of the sublists must match. If they don't maybe you want to concatenate the sublists along some axis. Try np.concatenate/np.r_/np.c_ etc. instead.

Converting list to numpy array

2 Answers2

Linked

Related