pythonic根据遮罩从3d数组中提取切片的方法

Question

I have an MxNxD array I and also a binary MxN mask M.

Let's say that there are k 1s in M. What I want is to extract a kxD array that contains all the D-length vectors corresponding to the 1s in the mask.

I can get the indices of these vectors in I by calling numpy.nonzero() but I can't find a nice compact way of getting my slice without horrible loops.

Any help will be much appreciated.

Answer 1

I think this is what you want:

In [283]: A = np.arange(24).reshape(2,3,4)
In [284]: M = np.array([[1,0,1],[0,1,0]],dtype=bool)
In [285]: A
Out[285]: 
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]]])
In [286]: M
Out[286]: 
array([[ True, False,  True],
       [False,  True, False]])
In [287]: I,J = np.nonzero(M)
In [288]: I,J
Out[288]: (array([0, 0, 1]), array([0, 2, 1]))
In [289]: A[I,J,:]
Out[289]: 
array([[ 0,  1,  2,  3],
       [ 8,  9, 10, 11],
       [16, 17, 18, 19]])

Since M is masking the initial dimensions, it can be simplified to

A[np.nonzero(M)]