I have a numpy matrix A and I need a function that will count (A[i,j]/sum of all elements in i-th column) - A[i,j]/sum of all elements in j-th row and it for every element in A.
I've already tried to write something along the line of:
a = A.sum(axis=0)
b = A.sum(axis=1)
for i in A:
for j in i:
f = A[i,j]/a(i) - A[i,j]/b(j)
You should probably look for a vectorized solution than looping through all elements
A/A.sum(axis=0)[np.newaxis,:] - A/A.sum(axis=1)[:,np.newaxis]
Suppose
In [35]: A = np.random.rand(10,3)
Out[35]:
array([[ 0.26070074, 0.58940996, 0.78665012],
[ 0.7420538 , 0.72214655, 0.66633183],
[ 0.67189673, 0.67298124, 0.04628626],
[ 0.93935375, 0.45030544, 0.38292913],
[ 0.45410731, 0.26557299, 0.09573014],
[ 0.99872912, 0.31092656, 0.46294278],
[ 0.61108329, 0.71140089, 0.85548017],
[ 0.80012964, 0.64749927, 0.3292407 ],
[ 0.33229818, 0.01810878, 0.44460486],
[ 0.86525557, 0.0569463 , 0.43183502]])
You could do the following.
In [36]: A*(1/A.sum(axis=0)[np.newaxis,:] - 1/A.sum(axis=1)[:,np.newaxis])
Out[36]:
array([[-0.12022572, -0.22751579, -0.3058817 ],
[-0.23713606, -0.17649948, -0.16474676],
[-0.3823249 , -0.33236228, -0.0229904 ],
[-0.38921906, -0.1527391 , -0.13097127],
[-0.48888156, -0.26594996, -0.09613741],
[-0.41381789, -0.10546217, -0.15833642],
[-0.18903571, -0.16660122, -0.20276794],
[-0.33044427, -0.21874513, -0.11216096],
[-0.36820095, -0.01870431, -0.46048657],
[-0.5094047 , -0.02924623, -0.22300407]])
I hope this could help:
def f(i,j,A):
return A[i,j] * ( 1. / np.sum(A[i,]) ) - ( 1. / np.sum(A[:,j]) )
