高效创建亲和矩阵numpy

Question

I currently use this code to compute an affinity matrix

m = np.size(pts[0])
A = np.zeros(m*m).reshape(m,m)
for i,x in enumerate(pts[0]):
    for j,y in enumerate(pts[1]):
        A[i,j] = np.exp(-1*np.linalg.norm(x-y)**2)

Where pts is a nx2 matrix. This works fine on smaller size but is O(n^2). The result should be a nxn matrix with the weights.

Answer 1

Here's a vectorized version:

#n = 10
#pts = np.random.rand(2, n)

X, Y = np.meshgrid(pts[1], pts[0])
A = np.exp(-np.abs(X-Y)**2)  # you can omit the abs because of **2