Welcome to MlosCore’s documentation!

_images/tests.svg_images/coverage.svg

mlos_core

This repository contains a stripped down implementation of essentially just the core optimizer and config space description APIs from the original MLOS.

It is intended to provide a simplified, easier to consume (e.g. via pip), with lower dependencies abstraction to

  • describe a space of context, parameters, their ranges, constraints, etc. and result objectives

  • an “optimizer” service abstraction (e.g. register() and suggest()) so we can easily swap out different implementations methods of searching (e.g. random, BO, etc.)

  • provide some helpers for automating optimization experiment runner loops and data collection

For these design requirements we intend to reuse as much from existing OSS libraries as possible and layer policies and optimizations specifically geared towards autotuning over top.

mlos_bench

This repository also contains the mlos_bench module intended to help automate and manage running experiments for autotuning systems with mlos_core.

See Also