Ax
Ax copied to clipboard
Adaptive Experimentation Platform
closes https://github.com/facebook/Ax/issues/2417
Summary: The current setup uses `node.model_spec_to_gen_from.model_key` to get the default name for a given node, which will attempt to fit models for GenNodes with multiple `model_spec`s. This diff avoids the...
Summary: This would previously filter out the observations with missing metrics with a warning, which could lead to issues in downstream usage. Returning the observations with NaN lets the downstream...
Summary: `get_outcome_constraint_transforms` evaluates the constraints by taking the product of tensor `A` with `Y` and comparing the outcome to `rhs` (uses einsum for this). The product of `0` and `nan`...
Summary: This avoids filtering out arms that are slightly outside the search space. Reviewed By: saitcakmak Differential Revision: D56936530
Summary: This diff enables multiple nodes to be used to generate a single batch trial. Right now the limitations are that: (1) currently each node only contributes 1 gr to...
Summary: In D56634321, observations_from_dataframe fails if there are metric_names Data.df that don't also exist on the experiment. This adjusts tests so that they avoid this issue. Differential Revision: D56850033
Summary: `Keys.PAIRWISE_PREFERENCE_QUERY` or `str(Keys.PAIRWISE_PREFERENCE_QUERY)` is commonly used when a string is expected, whereas I'm suspecting the intention is `Keys.PAIRWISE_PREFERENCE_QUERY.name`. This is causing issues in D56634321 which (for now) fails if...
The latest release removes the `GPKG` model, and suggests using the newer `BOTORCH_MODULAR` instead (https://github.com/facebook/Ax/pull/2316). However, I am not sure how to build a replacement for `GPKG` using this approach....
Summary: `PairwiseModelBridge` breaks if data passed to it include outcomes that are not preferences. We update it so that both supervised and ranking outcomes are supported. The updated bridge will...