Module math_util 

Mathematical utilities for distance computations and center finding.

This module contains optimized functions for computing squared L2 norms, finding closest centers, and processing residuals. These are primarily used in k-means clustering and disk index partitioning.

Functions

compute_closest_centers

Given data in num_points * new_dim row major Pivots stored in full_pivot_data as num_centers * new_dim row major Calculate the k closest pivot for each point and store it in vector closest_centers_ivf (row major, num_points*k) (which needs to be allocated outside) Additionally, if inverted index is not null (and pre-allocated), it will return inverted index for each center, assuming each of the inverted indices is an empty vector. Additionally, if pts_norms_squared is not null, then it will assume that point norms are pre-computed and use those values

compute_closest_centers_in_block

Calculate k closest centers to data of num_points * dim (row-major) Centers is num_centers * dim (row-major) data_l2sq has pre-computed squared norms of data centers_l2sq has pre-computed squared norms of centers Pre-allocated center_index will contain id of nearest center Pre-allocated dist_matrix should be num_points * num_centers and contain squared distances Default value of k is 1 Ideally used only by compute_closest_centers

compute_vecs_l2sq

Compute L2-squared norms of data stored in row-major num_points * dim, need to be pre-allocated