How to create groups based on an index array contains unequal shaped arrays

Question

I have an array and want to create groups based on an object-type like index array (which can get two different type forms, as that are shown under ids in the following code) e.g.:

poss = np.array([[-0.09060061464812283, -0.0686586855826326, -0.02647255439666547],
                 [0.08835853305001438, -0.06574591502086484, -0.05487544905792754],
                 [-0.06629393005890775, 0.00666283311876477, 0.06629393005890775],
                 [-0.06387415159186867, -0.1541029899066564, -0.03157952415698592],
                 [-0.0764878092204729, -0.17648780922047289, -0.04332722135612625]])

ids = np.array([[1, 3, 2], [4, 3, 1, 0]])
# <class 'numpy.ndarray'> <class 'list'> <class 'int'>
# <class 'numpy.ndarray'> <class 'numpy.ndarray'> <class 'numpy.int64'>

poss and ids are larger and in the above code they are just as an example. I know that if index array be as shape 2*4, it can be achieved by poss[ids, :]; But when the arrays in the index array have dissimilar shapes, what is the best way to create these groups Just by NumPy (preferably without looping)?

expected list/array:

# [[[ 0.08835853305001438 -0.06574591502086484 -0.05487544905792754]
#   [-0.06387415159186867 -0.1541029899066564  -0.03157952415698592]
#   [-0.06629393005890775  0.00666283311876477  0.06629393005890775]]
#
#  [[-0.0764878092204729  -0.17648780922047289 -0.04332722135612625]
#   [-0.06387415159186867 -0.1541029899066564  -0.03157952415698592]
#   [ 0.08835853305001438 -0.06574591502086484 -0.05487544905792754]
#   [-0.09060061464812283 -0.0686586855826326  -0.02647255439666547]]]

I don't think it's possible. Numpy has no operation to return an array containing arrays of different sizes. — Mechanic Pig
– Mechanic Pig, Commented May 22, 2022 at 15:23

mozway · Accepted Answer · 2022-05-22 15:24:55Z

1

This is not possible without at least looping on ids, as you have a ragged array (different sizes).

You can use:

out = np.split(poss[np.hstack(ids)], np.cumsum(list(map(len, ids)))[:-1])

output:

[array([[ 0.08835853, -0.06574592, -0.05487545],
        [-0.06387415, -0.15410299, -0.03157952],
        [-0.06629393,  0.00666283,  0.06629393]]),
 array([[-0.07648781, -0.17648781, -0.04332722],
        [-0.06387415, -0.15410299, -0.03157952],
        [ 0.08835853, -0.06574592, -0.05487545],
        [-0.09060061, -0.06865869, -0.02647255]])]

How it works:

stack ids as a single array a slice poss
split this array on the cumulated lengths of ids (here 3,7, the last split point being removed as this is the total length)

answered May 22, 2022 at 15:24

mozway

267k13 gold badges56 silver badges106 bronze badges

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

Collectives™ on Stack Overflow

How to create groups based on an index array contains unequal shaped arrays

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related