Tags: bletham/Ax
Tags
Pin BoTorch to 0.4.0 (facebook#507) Summary: Pull Request resolved: facebook#507 As discussed with Balandat offline yesterday, we should pin BoTorch versions so that we don't always have to release Ax right after BoTorch. Right now, we're in a slightly dangerous state where BoTorch 0.4.0 is released, and so because our setup.py file runs `pip install botorch>=0.3.3`, pip will install 0.4.0 now. But the stable version of Ax is actually not compatible with this newest version of BoTorch (as you can see from the failure of this test: https://github.com/facebook/Ax/runs/1967381927?check_suite_focus=true). Reviewed By: lena-kashtelyan Differential Revision: D26635518 fbshipit-source-id: bd34718a9faf55f3dbc0bf6115a96b72f4468372
Bump min botorch version (facebook#444) Summary: Pull Request resolved: facebook#444 To make the new contextual libraries happy, bump the botorch version (see failures in: https://fburl.com/kbkoen5s) Reviewed By: lena-kashtelyan Differential Revision: D25397593 fbshipit-source-id: c711ebb9d99477b5070d3758251d1d4a8b699b7a
remove sphinx-autodoc-typehints (facebook#419) Summary: Remove sphinx-autodoc-typehints from the sphinx/source/conf.py file. This was breaking our sphinx build, and is a temporary solution that will remove typehints from our docs. Pull Request resolved: facebook#419 Reviewed By: lena-kashtelyan Differential Revision: D24715038 Pulled By: Jakepodell fbshipit-source-id: 7133ab2b09f6b418488f475bf21497c19daebc3d
Add init to dir Summary: In facebook#387, users are having issue importing from the botorch_modular module. Maybe it's because of the missing __init__.py. But I have no clue how this works **without** it . . . Reviewed By: lena-kashtelyan Differential Revision: D24031280 fbshipit-source-id: 3cbfebf3f756927a78979941676d8c2aa34ac2f7
`ListSurrogate` Summary: `Surrogate` for `ModelListGP`: constructs a `ModelListGP` with specified sub-model classes under the hood Usage example: ``` m = BoTorchModel( surrogate=ListSurrogate( # Dictionary of which model class should be used to model which outcome. botorch_model_class_per_outcome={"m": SingleTaskGP, "n": SingleTaskGP} ) ) ``` D23606005 then introduces default usage of this for cases where there are multiple X-s in Xs and they are not all the same (so we are not using batched multi-output). Reviewed By: Balandat Differential Revision: D23382535 fbshipit-source-id: ff524c658e09a8837f71646f309289cde4da8c33
Remove unused variable in `choose_generation_strategy` Summary: As reported by an issue in github (facebook#341) This variable `no_max_parallelism` is not used by `choose_generation_strategy` but is still one of the functions input arguments. Cleaning it up in this diff Reviewed By: lena-kashtelyan Differential Revision: D22576370 fbshipit-source-id: 13c3d41b2c66151508a8d5e1290ff73b0601010f
update axis "type" (indicating log-scale) in interact_contour Summary: Interact contour plots certain param pairs incorrectly, choosing bad axis ranges. This fixes the problem. Reviewed By: Balandat Differential Revision: D21539918 fbshipit-source-id: b086a3469c75c21801471b3ab1069fb90abe4f3d
Expose properties column in SQATrial Summary: In MLHub, we use the properties column to store additional metadata on trials. We via access this via our custom MLHubDecoder, but we need to expose this column via SQA first. Reviewed By: lena-kashtelyan Differential Revision: D21064787 fbshipit-source-id: ba12059fdd08f7e93d6b19ee813831954b8f5f75
PreviousNext