How to make a numpy array from an array of arrays?

Question

I'm experimenting in ipython3, where I created an array of arrays:

In [105]: counts_array
Out[105]: 
array([array([ 17,  59, 320, ...,   1,   7,   0], dtype=uint32),
       array([ 30,  71, 390, ...,  12,  20,   6], dtype=uint32),
       array([  7, 145, 214, ...,   4,  12,   0], dtype=uint32),
       array([ 23, 346, 381, ...,  15,  19,   5], dtype=uint32),
       array([ 51,  78, 270, ...,   3,   0,   2], dtype=uint32),
       array([212, 149, 511, ...,  19,  31,   8], dtype=uint32)], dtype=object)

In [106]: counts_array.shape
Out[106]: (6,)

In [107]: counts_array[0].shape
Out[107]: (1590,)

I would like to obtain a plain shape=(6, 1590), dtype=uint32 array from this monster I created.

How can I do that?

Divakar · Accepted Answer · 2016-11-03 13:25:38Z

4

You can use np.vstack -

np.vstack(counts_array)

Another way with np.concatenate -

np.concatenate(counts_array).reshape(len(counts_array),-1)

Sample run -

In [23]: a
Out[23]: 
array([array([68, 92, 84, 35, 14, 71, 55, 40, 21, 41]),
       array([30, 90, 52, 64, 86, 68, 61, 85, 26, 98]),
       array([98, 64, 23, 49, 13, 17, 52, 96, 97, 19]),
       array([54, 26, 25, 22, 95, 77, 20, 73, 22, 80]),
       array([15, 84, 91, 54, 25, 21, 37, 19, 25, 25]),
       array([87, 17, 49, 74, 11, 34, 27, 23, 22, 83])], dtype=object)

In [24]: np.vstack(a)
Out[24]: 
array([[68, 92, 84, 35, 14, 71, 55, 40, 21, 41],
       [30, 90, 52, 64, 86, 68, 61, 85, 26, 98],
       [98, 64, 23, 49, 13, 17, 52, 96, 97, 19],
       [54, 26, 25, 22, 95, 77, 20, 73, 22, 80],
       [15, 84, 91, 54, 25, 21, 37, 19, 25, 25],
       [87, 17, 49, 74, 11, 34, 27, 23, 22, 83]])

answered Nov 3, 2016 at 13:25

Divakar

222k19 gold badges273 silver badges374 bronze badges

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

4 Comments

bli Over a year ago

Thanks. According to %timeit in ipython, on my data, your concatenate way is twice as fast as my array way. The vstack way is more than 4 .5 times slower.

Divakar Over a year ago

@bli Awesome ! Good to see some runtime numbers there! Not surprised by those, as vstack would internally use concatenate.

hpaulj Over a year ago

And vstack has to first reshape each array to 2d.

Divakar Over a year ago

@hpaulj That makes sense then for the timings, thanks!

bli · Accepted Answer · 2016-11-03 13:19:20Z

2

After various experiments, it turns out that the following simple syntax just works:

numpy.array([sub_array for sub_array in counts_array])

My first working version was unnecessary complicated:

numpy.array([[*sub_array] for sub_array in counts_array], dtype=numpy.uint32)

answered Nov 3, 2016 at 13:19

bli

8,3048 gold badges56 silver badges105 bronze badges

Comments

Miguel · Accepted Answer · 2016-11-03 13:28:49Z

2

Have you considered numpy.vstack()? I use it very often for this kind of operations.

answered Nov 3, 2016 at 13:28

Miguel

1592 silver badges10 bronze badges

Collectives™ on Stack Overflow

How to make a numpy array from an array of arrays?

3 Answers 3

4 Comments

Comments

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

3 Answers 3

4 Comments

Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Related