PerfTuned

Jump to bottom

Michael Ekstrand edited this page Nov 15, 2020 · 1 revision

This page is to collect a list of places where we're doing native-code optimization of LensKit code, so we can audit what's actually happening there and consider improved strategies.

Algorithms

Item-item model build (Numba parallel + MKL or SciPy sparse ops)
Item-item scoring
ALS model build (Numba parallel + BLAS-level NumPy ops
FunkSVD model build
User-user scoring
BPR sampling

Helpers

CSR matrix support
key-value heap (for item-item, maybe top-N)