问题
Suppose that I have a NumPy array arr like
array([5, 3, 0, 5, 7, 6, 5, 9, 0, 6])
and a few included indices in another array, inds,
array([3, 6])
I want to generate an array with the lengths of the subarrays of arr were I to split my array with inds. So in this case, my result would be [3, 3, 4].
I know that I could use np.split to do
>>> np.split(arr, inds)
[array([5, 3, 0]), array([5, 7, 6]), array([5, 9, 0, 6])]
and map size() to get the correct result of [3, 3, 4] from there, but it seems like an unnecessary cost to actually split the array given that I'm just using the size of the subarrays - this information is obviously available without splitting (from the indices), but how can I effectively use it?
回答1:
One approach with concatenation of the endpoints (0 and the length of array) on either sides of the indices array and then use diferentiation to get the interval lengths -
np.diff(np.concatenate(([0], inds, [arr.size])))
Shorter alternative -
np.diff(np.r_[0, inds, arr.size])
For performance we could use difference between one-off shifted slices to replace the differentiation with np.diff -
inds_ext = np.concatenate(([0], inds, [arr.size]))
out = inds_ext[1:] - inds_ext[:-1]
来源：https://stackoverflow.com/questions/42370491/getting-length-of-slices-of-numpy-array-or-list-without-actually-slicing