compress numpy array(matrix) by removing columns using another numpy array as mask

Question

I have a 2D numpy array (i.e matrix) A which contains useful data interspread with garbage in the form of column vectors as well as a 'selection' array B which contains '1' for those columns that are important and 0 for those that are not. Is there a way to select only those columns from A that correspond to ones in B? i.e i have a matrix

A = array([[ 0,  1,  2,  3,  4],   and a vector B = array([ 0,  1,  0,  1,  0])
           [ 5,  6,  7,  8,  9],
           [10, 11, 12, 13, 14],
           [15, 16, 17, 18, 19],
           [20, 21, 22, 23, 24]])

and I want

array([[1,   3],
       [6,   8],
       [11, 13],
       [16, 18],
       [21, 23]])

Is there an elegant way to do so? Right now i just have a for loop that iterates through B.

NOTE: the matrices that i'm dealing with are large, so i don't want to use numpy masked arrays, as i simply don't want the masked data

Answer 1

>>> A
  array([[ 0,  1,  2,  3,  4],
         [ 5,  6,  7,  8,  9],
         [10, 11, 12, 13, 14],
         [15, 16, 17, 18, 19],
         [20, 21, 22, 23, 24]])
>>> B = NP.array([ 0,  1,  0,  1,  0])

>>> # convert the indexing array to a boolean array
>>> B = NP.array(B, dtype=bool)

>>> # index A against B--indexing array is placed after the ',' because
>>> # you are selecting columns

>>> res = A[:,B]

>>> res
  array([[ 1,  3],
         [ 6,  8],
         [11, 13],
         [16, 18],
         [21, 23]])  

The syntax for index-based slicing in NumPy is elegant and simple. A couple of rules cover a majority of use cases:

• the form is [rows, columns]

• specify all rows or all columns using a colon ":" e.g., [:, 4] (extracts the entire 5th column)

Answer 2

Not sure if it's the most efficient way (because of the transposition), but it should be better than a for loop:

A.T[B == 1].T