Use case

Compression, quantization, and indexing in vector search databases.

Describe the solution you'd like

L2NearestCentroid(vec, centroids)

Takes vec - Array of numbers, typically Float32,
and centroids - constant Array of Arrays - vectors to compare.

Returns the index of the nearest centroid, according to the L2 distance.
For convenience, the index can count from zero instead of one.

partitionByHyperplanes(vec, normals, [offsets])

Takes vec - Array of numbers, typically Float32, representing a point in vector space,
normals - constant Array of Arrays, each of them representing a normal vector for a hyperplane,
offsets (optional) - constant Array of Arrays, each of them representing an offset for the corresponding hyperplane (default zero).

Note: every hyperplane has the following equation (vec - offset) ⋅ normal = 0.

Returns FixedString with every bit corresponding to a subspace, relative to the corresponding hyperplane.

Additional context

These are one of the few functions, that could benefit from GPU.