How to group elements of a numpy array with the same value in separate numpy arrays

Question

As usual intro, I am a tyro in python. However, I got quite a big project to code. It is a surface flow model with Cell Automata. Anyway, I also want to include building roofs in my model. Imagine you have an ascii file indicating buildings with 1s, while the rest is 0. There are just those two states. Now, I want to find all adjacent cells indicating the same building and store them (or rather the information of y,x and one more (maybe elevation),so 3 columns) in an individual building arrays. Keep in mind that buildings can have all possible forms though diagonally connected cells doesn't belong to the same building. So only northern, southern, western and eastern cells can belong to the same building.

I did my homework and googled it but so far I couldn't find a satisfying answer.

example: initial land-cover array:

([0,0,0,0,0,0,0]
 [0,0,1,0,0,0,0]
 [0,1,1,1,0,1,1]
 [0,1,0,1,0,0,1]
 [0,0,0,0,0,0,0])

output(I need to now the coordinates of the cells in my initial array):

 building_1=([1,2],[2,1],[2,2],[2,3],[3,1],[3,3])
 building_2=([2,5],[2,6],[3,6])

Any help is greatly appreciated!

Answer 1

You can use the label function from scipy.ndimage to identify the distinct buildings.

Here's your example array, containing two buildings:

In [57]: a
Out[57]: 
array([[0, 0, 0, 0, 0, 0, 0],
       [0, 0, 1, 0, 0, 0, 0],
       [0, 1, 1, 1, 0, 1, 1],
       [0, 1, 0, 1, 0, 0, 1],
       [0, 0, 0, 0, 0, 0, 0]])

Import label.

In [58]: from scipy.ndimage import label

Apply label to a. It returns two values: the array of labeled positions, and the number of distinct objects (buildings, in this case) found.

In [59]: lbl, nlbls = label(a)

In [60]: lbl
Out[60]: 
array([[0, 0, 0, 0, 0, 0, 0],
       [0, 0, 1, 0, 0, 0, 0],
       [0, 1, 1, 1, 0, 2, 2],
       [0, 1, 0, 1, 0, 0, 2],
       [0, 0, 0, 0, 0, 0, 0]], dtype=int32)

In [61]: nlbls
Out[61]: 2

To get the coordinates of a building, np.where can be used. For example,

In [64]: np.where(lbl == 2)
Out[64]: (array([2, 2, 3]), array([5, 6, 6]))

It returns a tuple of arrays; the kth array holds the coordinates of the kth dimension. You can use, for example, np.column_stack to combine these into an array:

 In [65]: np.column_stack(np.where(lbl == 2))
 Out[65]: 
 array([[2, 5],
        [2, 6],
        [3, 6]])

You might want a list of all the coordinate arrays. Here's one way to create such a list.

For convenience, first create a list of labels:

In [66]: labels = range(1, nlbls+1)

In [67]: labels
Out[67]: [1, 2]

Use a list comprehension to create the list of coordinate arrays.

In [68]: coords = [np.column_stack(where(lbl == k)) for k in labels]

In [69]: coords
Out[69]: 
[array([[1, 2],
       [2, 1],
       [2, 2],
       [2, 3],
       [3, 1],
       [3, 3]]),
 array([[2, 5],
       [2, 6],
       [3, 6]])]

Now your building data is in labels and coords. For example, the first building was labeled labels[0], and its coordinates are in coords[0]:

In [70]: labels[0]
Out[70]: 1

In [71]: coords[0]
Out[71]: 
array([[1, 2],
       [2, 1],
       [2, 2],
       [2, 3],
       [3, 1],
       [3, 3]])