问题
I want to perform some numpy methods on items of structured array instead of numbers. So, for example, while working with array of integers of shape (4, 3), I need to convert it to array of items of length 3 and perform some operations as it were a single one dimensional array of shape (4,). These conversions itself, unfortunately, looks really complicated for me. Let's take another example:
n, m, r = 2, 3, 4
array = np.arange(n*m).reshape((n,m))
dt = np.dtype(','.join('i'*m))
arr1 = np.array([tuple(x) for x in array], dtype=dt)
>>> arr1
array([(0, 1, 2), (3, 4, 5)],
dtype=[('f0', '<i4'), ('f1', '<i4'), ('f2', '<i4')])
Then I call some methods on it, let it for the sake of simplicity be np.tile
(but they might be completely different):
arr2 = np.tile(arr1[:,None], r)
>>> arr2
array([[(0, 1, 2), (0, 1, 2), (0, 1, 2), (0, 1, 2)],
[(3, 4, 5), (3, 4, 5), (3, 4, 5), (3, 4, 5)]],
dtype=[('f0', '<i4'), ('f1', '<i4'), ('f2', '<i4')])
I want to convert it into this array instead:
array([[[0, 1, 2], [0, 1, 2], [0, 1, 2], [0, 1, 2]],
[[3, 4, 5], [3, 4, 5], [3, 4, 5], [3, 4, 5]]]
I have two questions:
- How to convert
array
toarr1
(a bunch of special items) without iteration? - How to convert
arr2
(a bunch of special items) back to array of single items I want?
回答1:
numpy
provides a helper functions to do this:
>>> n, m, r = 2, 3, 4
>>> array = np.arange(n*m).reshape((n,m))
>>> import numpy.lib.recfunctions as recfunctions
>>> recfunctions.unstructured_to_structured(array, dtype=np.dtype(','.join('i'*m)))
array([(0, 1, 2), (3, 4, 5)],
dtype=[('f0', '<i4'), ('f1', '<i4'), ('f2', '<i4')])
And in the other direction:
>>> import numpy.lib.recfunctions as recfunctions
>>> recfunctions.structured_to_unstructured(arr2)
array([[[0, 1, 2],
[0, 1, 2],
[0, 1, 2],
[0, 1, 2]],
[[3, 4, 5],
[3, 4, 5],
[3, 4, 5],
[3, 4, 5]]], dtype=int32)
In this particular case, if the original array is dtype=np.int32
, you could use a view:
>>> array = np.arange(n*m, dtype=np.int32).reshape((n,m))
>>> structured_view = array.view(dtype=np.dtype(','.join('i'*m)))
>>> structured_view
array([[(0, 1, 2)],
[(3, 4, 5)]], dtype=[('f0', '<i4'), ('f1', '<i4'), ('f2', '<i4')])
The advantage of a view is that it create a new array. Of course, this can be a disadvantage if you mutate your view and don't expect the original array to change as well.
In the reverse, it doesn't handle the shape you want, but you could always reshape:
>>> arr2.view(dtype=np.int32)
array([[0, 1, 2, 0, 1, 2, 0, 1, 2, 0, 1, 2],
[3, 4, 5, 3, 4, 5, 3, 4, 5, 3, 4, 5]], dtype=int32)
Using views can get tricky, fast.
来源:https://stackoverflow.com/questions/64613276/how-to-convert-array-into-special-items-of-structured-array-and-revert-it-back