Convert 1D object numpy array of lists to 2D numeric array and back

Question

Say I have this object array containing lists of the same length:

>>> a = np.empty(2, dtype=object)
>>> a[0] = [1, 2, 3, 4]
>>> a[1] = [5, 6, 7, 8]
>>> a
array([[1, 2, 3, 4], [5, 6, 7, 8]], dtype=object)

How can I convert this to a numeric 2D array?

>>> a.shape
(2,)
>>> b = WHAT_GOES_HERE(a)
>>> b
array([[1, 2, 3, 4],
       [5, 6, 7, 8]])
>>> b.shape
(2, 4)

How can I do the reverse?

Does it get easier if my a array is an np.array of np.arrays, rather than an np.array of lists?

>>> na = np.empty(2, dtype=object)
>>> na[0] = np.array([1, 2, 3, 4])
>>> na[1] = np.array([5, 6, 7, 8])
>>> na
array([array([1, 2, 3, 4]), ([5, 6, 7, 8])], dtype=object)

@Divakar: ValueError: setting an array element with a sequence. - astype can't reshape — Eric
– Eric, Commented Sep 22, 2016 at 11:16

Divakar · Accepted Answer · 2016-09-22 11:56:53Z

3

One approach using np.concatenate -

b = np.concatenate(a).reshape(len(a),*np.shape(a[0]))

The improvement suggest by @Eric to use *np.shape(a[0]) should make it work for generic ND shapes.

Sample run -

In [183]: a
Out[183]: array([[1, 2, 3, 4], [5, 6, 7, 8]], dtype=object)

In [184]: a.shape
Out[184]: (2,)

In [185]: b = np.concatenate(a).reshape(len(a),*np.shape(a[0]))

In [186]: b
Out[186]: 
array([[1, 2, 3, 4],
       [5, 6, 7, 8]])

In [187]: b.shape
Out[187]: (2, 4)

To get back a, it seems we can use a two-step process, like so -

a_back = np.empty(b.shape[0], dtype=object)
a_back[:] = b.tolist()

Sample run -

In [190]: a_back = np.empty(b.shape[0], dtype=object)
     ...: a_back[:] = b.tolist()
     ...: 

In [191]: a_back
Out[191]: array([[1, 2, 3, 4], [5, 6, 7, 8]], dtype=object)

In [192]: a_back.shape
Out[192]: (2,)

edited Sep 22, 2016 at 11:56

answered Sep 22, 2016 at 11:19

Divakar

222k19 gold badges273 silver badges374 bronze badges

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

3 Comments

Eric Over a year ago

Nice concise solution for the reverse - I was thinking I might have to resort to a for loop

Eric Over a year ago

Replacing the -1 with np.shape(a[0]) makes this work for the more general 1D of ND -> N+1D case too

Divakar Over a year ago

@Eric Thanks for the improvement! Updated the post.

Johannes Hinrichs · Accepted Answer · 2016-09-22 11:22:12Z

2

You canuse np.vstack():

>>> a = np.vstack(a).astype(int)

edited Sep 22, 2016 at 11:22

answered Sep 22, 2016 at 11:20

Johannes Hinrichs

17112 bronze badges

3 Comments

Eric Over a year ago

I think the astype() is unecessary too!

Johannes Hinrichs Over a year ago

yes, you are right, although you didn't explicitly say you wanted int in the end (although - what esle in this example...), so better leave it in ;)

Eric Over a year ago

I just said numeric, and in this case the dtype is int32 anyway, it seems

mtrw · Accepted Answer · 2016-09-22 11:22:20Z

0

Here's an approach that converts the source NumPy array to lists and then into the desired NumPy array:

b = np.array([k for k in a])
b
array([[1, 2, 3, 4],
       [5, 6, 7, 8]])
c = np.array([k for k in b], dtype=object)
c
array([[1, 2, 3, 4],
       [5, 6, 7, 8]], dtype=object)

answered Sep 22, 2016 at 11:22

mtrw

35.3k7 gold badges66 silver badges73 bronze badges

Comments

Eric · Accepted Answer · 2016-09-22 11:32:19Z

0

I found that round-tripping through list with np.array(list(a)) was sufficient.

This seems to be equivalent to using np.stack(a).

Both of these have the benefit of working in the more general case of converting a 1D array of ND arrays into an (N+1)D array.

edited Sep 22, 2016 at 11:32

answered Sep 22, 2016 at 11:22

Eric

98.1k54 gold badges257 silver badges389 bronze badges

Collectives™ on Stack Overflow

Convert 1D object numpy array of lists to 2D numeric array and back

4 Answers 4

3 Comments

3 Comments

Comments

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

4 Answers 4

3 Comments

3 Comments

Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Related