MLOS
MLOS copied to clipboard
microsoft
MLOS is a project to enable autotuning for systems.
```suggestion assert os.path.samefile(local_exec_service._temp_dir, path_join(temp_dir, "temp", abs_path=True)) ``` or something like it? + We should probably check for all other places where we compare paths _Originally posted by @motus in https://github.com/microsoft/MLOS/pull/531#discussion_r1352961180_
Experiment's can often be related (e.g., by only changing a handful of tunable parameters). In that case, we may wish to reuse some of the resources used for a group...
- [x] unify json schema to a single top level schema - Idea is that it just selects subschemas based on class prefix e.g., `"class": "mlos_bench.service.*" maps to --> "service-schema.json"`,...
- [ ] Test integer keyed parameters (requires additional fixes) - [ ] Test param names with special characters (e.g., `/`)
> Might just need to verify that we can reprime and finish an optimizer loop without that directory around. Not sure if we have a test for that yet ......
- [ ] should include a practical description of what the project enables and how it differs from existing solutions - [x] ~should include simple instructions on getting started with...
Some systems have constraints that our optimizers don't currently model (e.g. param A must be a linear combination of param B). When set to something else, they actually run with...
Right now, if a trial fails, we simply continue. This seems reasonable for benchmark environment failures, however, for lower level failures (e.g. VM, OS) that the leaf environments rely upon...
Per #528 > Future work: add support for remote exec version of this
It might be nice if the path resolution logic for a config allowed for paths relative to that config file (e.g. to reference scripts within that directory) as opposed to...