Creating a numpy matrix from a 1dim numpy array by performing computation on itself?

Question

I have a numpy array 'arr' of shape (100000,). I need to create a numpy matrix 'res_matrix' of shape 100000X100000 such that

for i in range(res_matrix.shape[0]):
    for j in range(res_matrix.shape[1]):
        res_matrix[i][j]= arr[i]*arr[j]

Sample input/output

arr=[1 2 4]


Output:
res_matrix:

[[1 2 4]
 [2 4 18]
 [4 8 16]]

Is there way to do vectorize this operation, to reduce the computation time of calculating 00000X100000 in loop?

Answer 1

There are a few ways you can get an outer multiplication.

arr = np.array([1,2,4])

#Using Multiply outer
print(np.multiply.outer(arr, arr)) #As suggested by Warren

#Using broadcasting
print(arr[:,None] * arr[None,:]) #(3,1) * (1,3)

[[ 1  2  4]
 [ 2  4  8]
 [ 4  8 16]]

[[ 1  2  4]
 [ 2  4  8]
 [ 4  8 16]]

​ Note, the output is still a very large matrix for storing in memory. Depending on what you need it for, I would advise considering something like a generator function. Let me know how you are going to use this matrix and I could suggest more memory efficient methods.