numpy.arange on an N x 2 array-CodePudding

Given an (N, 2) array of [start, stop] values, I want an (M,) array consisting of [np.arange(start[0], stop[0]), np.arange(start[1], stop[1], ..., np.arange(start[N - 1], stop[N - 1])].

For example, given arr = np.array([[1, 3], [4, 5], [7, 8], [8, 10]]), I want [1, 2, 4, 7, 8, 9].

One method is with a for loop:

arr = np.array([[1, 3], [4, 5], [7, 8], [8, 10]])
out = np.zeros(0, dtype=int)
for i in range(len(arr)):
  out = np.append(out, np.arange(arr[i, 0], arr[i, 1]))
print(out)

Result: [1 2 4 7 8 9].

Is there a more efficient and/or readable way of doing this?

CodePudding user response：

Using a list comprehension to collect the aranges, and then join them:

In [214]: alist = [[1, 3], [4, 5], [7, 8], [8, 10]]
In [215]: [np.arange(start,stop) for start,stop in alist]
Out[215]: [array([1, 2]), array([4]), array([7]), array([8, 9])]
In [216]: np.hstack(_)
Out[216]: array([1, 2, 4, 7, 8, 9])

There's a version of hstack that uses indexing syntax:

In [217]: np.r_[1:3, 4:5, 7:8, 8:10]
Out[217]: array([1, 2, 4, 7, 8, 9])

It translates the slices into arange calls. It's compact, but it's not easy to use with a predefined list.

With the predefined list, first create a tuple:

In [225]: tup = tuple([slice(start,stop) for start, stop in alist])
In [226]: tup
Out[226]: (slice(1, 3, None), slice(4, 5, None), slice(7, 8, None), slice(8, 10, None))
In [227]: np.r_[tup]
Out[227]: array([1, 2, 4, 7, 8, 9])

Anyways, there isn't a way of generating multiple aranges with one call. If the aranges are all the same length it is possible to use np.linspace, but not when they very in length.

We try to avoid repeated np.append. It is inferior to list append.

CodePudding user response：

Here the line you need.

[item for start, stop in [[1, 3], [4, 5], [7, 8], [8, 10]] for item in range(start, stop)]

You can deal with normal list to achieve this. (Don t overuse numpy, python vanilla is good too!)

For small amount of data, the numpy import don't worth it!

Python vanilla time for this problem: 2.5987625122070312e-05 s
Your method time for this problem: 0.09107470512390137 s
Same without numpy import: 0.00015807151794433594 s