Spaces:
Build error
Build error
| import numpy as np | |
| from numba import jit, prange | |
| # https://github.com/talboger/fastdist | |
| def euclidean_vector_to_matrix_distance(u, m): | |
| """ | |
| :purpose: | |
| Computes the distance between a vector and the rows of a matrix using any given metric | |
| :params: | |
| u : input vector of shape (n,) | |
| m : input matrix of shape (m, n) | |
| distance vector : np.array, of shape (m,) vector containing the distance between u | |
| and the rows of m | |
| :example: | |
| >>> from fastdist import fastdist | |
| >>> import numpy as np | |
| >>> u = np.random.RandomState(seed=0).rand(10) | |
| >>> m = np.random.RandomState(seed=0).rand(100, 10) | |
| >>> fastdist.vector_to_matrix_distance(u, m) | |
| (returns an array of shape (100,)) | |
| :note: | |
| the cosine similarity uses its own function, cosine_vector_to_matrix. | |
| this is because normalizing the rows and then taking the dot product | |
| of the vector and matrix heavily optimizes the computation. the other similarity | |
| metrics do not have such an optimization, so we loop through them | |
| """ | |
| n = m.shape[0] | |
| out = np.zeros((n), dtype=np.float32) | |
| for i in prange(n): | |
| dist = 0 | |
| for l in range(len(u)): | |
| dist += abs(u[l] - m[i][l]) ** 2 | |
| out[i] = dist ** (1 / 2) | |
| return out | |