检查矩阵/数组的索引，然后检查某些条件

Question

This code converts a square matrix to strict lower triangular matrix (lower elements to 0)

a=np.random.randn(9).reshape((3,3))

a
Out[61]: 
array([[-0.18314209,  0.3710528 , -1.46067261],
       [-0.55834476, -1.41924213, -0.04127718],
       [ 0.40134248, -0.41759044,  1.83573994]])

def subs_tri_0(mat,i,j): mat[i,j] = 0

[subs_tri_0(a,i,j) for i,j in product(xrange(a.shape[0]),xrange(a.shape[1])) if i > j]
Out[63]: [None, None, None]

a
Out[64]: 
array([[-0.18314209,  0.3710528 , -1.46067261],
       [ 0.        , -1.41924213, -0.04127718],
       [ 0.        ,  0.        ,  1.83573994]])

Is there a way to this using short and sweet where?

Answer 1

It doesn't use numpy.where , but you could use numpy.tril_indices to set the lower triangular to zero:

>>> a
array([[ 0.05559341, -1.93583316, -1.19666435],
       [-0.33450047,  0.63275874,  0.77152195],
       [-0.73106122, -1.57602057,  0.41878224]])
>>> a[np.tril_indices(3, k=-1)] = 0
>>> a
array([[ 0.05559341, -1.93583316, -1.19666435],
       [ 0.        ,  0.63275874,  0.77152195],
       [ 0.        ,  0.        ,  0.41878224]])

Note that you need to pass k=-1 to numpy.tril_indices to not include the diagonal.

Answer 2

Found the result. Although I'm gona go ahead with @mdml solution. which is to the point btw.

This is just a positive side effect of looping through indices to check for row,col conditions. perhaps anyone could use it for some other meaningful purpose where tril and triu methods are not the case.

a=np.random.randn(9).reshape((3,3))

r,c = where(a)
where((r>c).reshape(a.shape),0,a)
Out[169]: 
array([[ 1.49230558,  0.6321149 ,  0.05299907],
       [ 0.        ,  0.14736346,  0.42516369],
       [ 0.        ,  0.        , -0.6878655 ]])