I have a large numpy array (typically a few thousands of numbers) that is consisted of several sorted sequences,
for example:
arr = [12, 13, 14, 22, 23, 24, 25, 26, 9, 10, 11]
I would like to split it into subarrays - each one holds another sequence -
[12, 13, 14], [22, 23, 24, 25, 26], [9, 10, 11]
What is the fastest way to do that?
I would do it following way
import numpy as np
arr = np.array([12, 13, 14, 22, 23, 24, 25, 26, 9, 10, 11])
splits = np.flatnonzero(np.diff(arr)!=1)
sub_arrs = np.split(arr, splits+1)
print(sub_arrs)
output
[array([12, 13, 14]), array([22, 23, 24, 25, 26]), array([ 9, 10, 11])]
Explanation: I create array with differences between adjacent elements using numpy.diff (np.diff(arr)) then process it to get array with Trues where difference is 1 and Falses in every other case (np.diff(arr)!=1) then find indices of Trues in that array using np.flatnonzero (True is treated as 1 and False is treated as 0 in python) finally I use numpy.split to get list of subarrays made from arr at spllited at splits offseted by 1 (note that numpy.diff returns array which is shorter by 1 than its input).
Side note: I would call this finding sub-arrays with consecutive runs, rather than merely sorted as you might split your arr into [[12, 13, 14, 22, 23, 24, 25, 26], [9, 10, 11]] and full-fill requirement that every sub-array is sorted
First of all, the problem could be really complex, but based on your example I assume that the values in subarrays are increasing by 1.
Here is a one liner solution with plain numpy: np.array_split(a, np.where(np.diff(a) != 1)[0]+1)
Explanation: You can calculate the difference between consecutive values with np.diff.
>>> import numpy as np
>>> a
array([12, 13, 14, 22, 23, 24, 25, 26, 9, 10, 11])
>>> np.diff(a)
array([ 1, 1, 8, 1, 1, 1, 1, -17, 1, 1])
Then, get the indices of the values that represents the last element of the subarrays, that is the values that do no equal 1.
>>> np.where(np.diff(a) != 1)
(array([2, 7]),)
Finally, we add 1 to the boundaries to be able to use np.array_split() correctly to generate the subarrays.
>>> np.where(np.diff(a) != 1)[0]+1
array([3, 8])
>>> np.array_split(a, np.where(np.diff(a) != 1)[0]+1)
[array([12, 13, 14]), array([22, 23, 24, 25, 26]), array([ 9, 10, 11])]
