Indexing different sized ranges in a 2D numpy array using a Pythonic vectorized code

Question

I have a numpy 2D array, and I would like to select different sized ranges of this array, depending on the column index. Here is the input array a = np.reshape(np.array(range(15)), (5, 3)) example

[[ 0  1  2]
 [ 3  4  5]
 [ 6  7  8]
 [ 9 10 11]
 [12 13 14]]

Then, list b = [4,3,1] determines the different range sizes for each column slice, so that we would get the arrays

[0 3 6 9]
[1 4 7]
[2]

which we can concatenate and flatten to get the final desired output

[0 3 6 9 1 4 7 2]

Currently, to perform this task, I am using the following code

slices = []
for i in range(a.shape[1]):
    slices.append(a[:b[i],i])

c = np.concatenate(slices)

and, if possible, I want to convert it to a pythonic format.

Bonus: The same question but now considering that b determines row slices instead of columns.

Answer 1

We can use broadcasting to generate an appropriate mask and then masking does the job -

In [150]: a
Out[150]: 
array([[ 0,  1,  2],
       [ 3,  4,  5],
       [ 6,  7,  8],
       [ 9, 10, 11],
       [12, 13, 14]])

In [151]: b
Out[151]: [4, 3, 1]

In [152]: mask = np.arange(len(a))[:,None] < b

In [153]: a.T[mask.T]
Out[153]: array([0, 3, 6, 9, 1, 4, 7, 2])

Another way to mask would be -

In [156]: a.T[np.greater.outer(b, np.arange(len(a)))]
Out[156]: array([0, 3, 6, 9, 1, 4, 7, 2])

Bonus : Slice per row

If we are required to slice per row based on chunk sizes, we would need to modify few things -

In [51]: a
Out[51]: 
array([[ 0,  1,  2,  3,  4],
       [ 5,  6,  7,  8,  9],
       [10, 11, 12, 13, 14]])

# slice lengths per row
In [52]: b
Out[52]: [4, 3, 1]

# Usual loop based solution :
In [53]: np.concatenate([a[i,:b_i] for i,b_i in enumerate(b)])
Out[53]: array([ 0,  1,  2,  3,  5,  6,  7, 10])

# Vectorized mask based solution :
In [54]: a[np.greater.outer(b, np.arange(a.shape[1]))]
Out[54]: array([ 0,  1,  2,  3,  5,  6,  7, 10])