pairwise-distance-matrix — Pinterest case

Pinterest-inspired visual retrieval prototype compares a batch of item embeddings against every other item in the batch. You need to compute the pairwise squared-distance matrix efficiently and with the expected shape.

Compute the matrix of squared Euclidean distances between all pairs of rows of X: entry $(i,j)(i,j)$ is $\mathrm{\parallel }{x}_i-x_j{\mathrm{\parallel }}^2\backslash |x\_i-x\_j\backslash |^2$.

Implement solve(X).

Returns. Return a symmetric $n\times nn\backslash times\; n$ float NumPy matrix with zeros on the diagonal.

Examples

Example 1

Input:

solve([[0, 0], [3, 4]])

Output:

[[0.0, 25.0], [25.0, 0.0]]

Example 2

Input:

solve([[1.0], [2.0], [4.0]])

Output:

[[0.0, 1.0, 9.0], [1.0, 0.0, 4.0], [9.0, 4.0, 0.0]]