Convert elements of multi-dimensional numpy array to float32 - python-3.x

Question

I have a complicated nested numpy array which contains list. I am trying to converted the elements to float32. However, it gives me following error:
ValueError Traceback (most recent call last)
<ipython-input-225-22d2824961c2> in <module>
----> 1 x_train_single.astype(np.float32)
ValueError: setting an array element with a sequence.
Here is the code and sample input:
x_train_single.astype(np.float32)
array([[ list([[[0, 0, 0, 0, 0, 0]], [-1.0], [0]]),
list([[[0, 0, 0, 0, 0, 0], [173, 8, 172, 0, 0, 0]], [-1.0], [0]])
]])

Answer 1

As your array contains lists of different sizes and nesting depths, I doubt that there is a simple or fast solution.
Here is a "get-the-job-done-no-matter-what" approach. It comes in two flavors. One creates arrays for leaves, the other one lists.
>>> a
array([[list([[[0, 0, 0, 0, 0, 0]], [-1.0], [0]]),
list([[[0, 0, 0, 0, 0, 0], [173, 8, 172, 0, 0, 0]], [-1.0], [0]])]],
dtype=object)
>>> def mkarr(a):
... try:
... return np.array(a,np.float32)
... except:
... return [*map(mkarr,a)]
...
>>> def mklst(a):
... try:
... return [*map(mklst,a)]
... except:
... return np.float32(a)
...
>>> np.frompyfunc(mkarr,1,1)(a)
array([[list([array([[0., 0., 0., 0., 0., 0.]], dtype=float32), array([-1.], dtype=float32), array([0.], dtype=float32)]),
list([array([[ 0., 0., 0., 0., 0., 0.],
[173., 8., 172., 0., 0., 0.]], dtype=float32), array([-1.], dtype=float32), array([0.], dtype=float32)])]],
dtype=object)
>>> np.frompyfunc(mklst,1,1)(a)
array([[list([[[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]], [-1.0], [0.0]]),
list([[[0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [173.0, 8.0, 172.0, 0.0, 0.0, 0.0]], [-1.0], [0.0]])]],
dtype=object)

Answer 2

if number of columns is fixed then
np.array([l.astype(np.float) for l in x_train_single.squeeze()])
But it will remove the redundant dimensions, convert everything into numpy array.
Before: (1, 1, 1, 11, 6)
After: (11,6)

Answer 3

Try this:
np.array(x_train_single.tolist())
Looks like you have a (1,1) shaped array, where the single element is a list. And the sublists a consistent in size.
I expect you will get an array with shape (1, 1, 1, 11, 6) and int dtype.
or:
np.array(x_train_single[0,0])
Again this extracts the list from the array, and then makes an array from that.
My answer so far was based on the display:
array([[list([[[173, 8, 172, 0, 0, 0], [512, 58, 57, 0, 0, 0],
...: [513, 514, 0, 0, 0, 0], [515, 189, 516, 0, 0, 0], [309, 266, 0, 0, 0,
...: 0],
...: [32, 310, 0, 0, 0, 0], [271, 58, 517, 0, 0, 0], [164, 40, 0, 0, 0, 0],
...: [38, 32, 60, 0, 0, 0], [38, 83, 60, 0, 0, 0], [149, 311, 0, 0, 0, 0]]
...: ])]])
The new display is more complicated
array([[ list([[[0, 0, 0, 0, 0, 0]], [-1.0], [0]]),
...: list([[[0, 0, 0, 0, 0, 0], [173, 8, 172, 0, 0, 0]], [-1.0], [0]])]])
because the inner lists differ in size. It can't be made into a numeric dtype array.
It can be turned into a (1,2,3) shape array, but still object dtype with 1d list elements.