Docusaurus versions 1.1.0 #4136
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Summary: Pull Request resolved: #3851 Introduces `ExperimentData`, a container of two dataframes that is intended to replace the use of lists of `Observation` objects within the `Adapter` and `Transform` layers. By replacing lists of objects with dataframes, we enable vectorized operations (replacing many nested for loops), which results in 4-5x speed up in data processing within Adapter. arm_data: A dataframe containing columns for trial index, arm name, and the parameterization of each arm. Each row corresponds to a (trial_index, arm_name) pair. observation_data: A dataframe containing columns for trial index, arm name, any map keys (e.g., progression / step), and the mean and sem observations for the metrics. Each row corresponds to an observation for the given trial index, arm name, and map key values / progression. This is typically constructed by pivoting `(Map)Data.true_df`. Reviewed By: esantorella Differential Revision: D69943574 fbshipit-source-id: 26a9f77a2d8ef58b7bffa3ac37ffb5e317bd9b88
Summary: Pull Request resolved: #3858 `OneHot` transform had a duplicate `_` in the transformed parameter names. This diff removes that. With this change, the transformed parameter name will be - For two values only: `parameter_OH_PARAM`, which was `parameter_OH_PARAM_` - For more values: `parameter_OH_PARAM_i` which was `parameter_OH_PARAM__i`. Reviewed By: Balandat Differential Revision: D75892920 fbshipit-source-id: f5980968dca45ab7403af913b371c5a35334aa29
Summary: Pull Request resolved: #3829 Reviewed By: lena-kashtelyan Differential Revision: D74902347 fbshipit-source-id: 9876ab3d27c040130ca4b3b0acf85a2e6b531747
Summary: Pull Request resolved: #3852 This diff adds a `metadata` column to `arm_data` of `ExperimentData`, which was introduced in the previous diff. The `metadata` is extracted from the `GeneratorRun.candidate_metadata_by_arm_signature` (thus available at `(trial_index, arm_name)` level, same as the index of `arm_data`), and is traditionally present as `ObservationFeatures.metadata`. This is extracted in `TorchAdapter._fit` and passed down to `Generator.fit` (only to be used by TRBO AFAICT). Alternative: We could add metadata as a separate dataframe on `ExperimentData` but I don't see a clear benefit from doing so. - Pros: It'd eliminate the need to exclude `metadata` column when accessing only the parameterizations from `arm_data`, which we need to do in a few places. - Cons: When extracting the metadata in `Adapter.fit`, we would have additional indexing operations to ensure that the `metadata` matches the index of `arm_data` and `observation_data`. Reviewed By: esantorella Differential Revision: D72801799 fbshipit-source-id: 22339f159e3469bee399a8760beb5ff1941e5307
Summary: Making sure that the torch_device is set to the same as the model in the client.compute_analyses in case CUDA is used in the Modular BoTorch Interface. Related to the issue [https://github.com/facebook/Ax/issues/3813](https://github.com/facebook/Ax/issues/3813) and without the messed-up formatting from #3833, following Balandat's advice. Pull Request resolved: #3839 Reviewed By: hvarfner Differential Revision: D75758870 Pulled By: Balandat fbshipit-source-id: bec3a7c51219a334192db390f1ed93d4d9571929
Summary: Pull Request resolved: #3854 Update `ArmEffectsPlot` and `ScatterPlot` to accept all `TrialStatus`'s that are expecting data, not just `COMPLETED`. Reviewed By: sdaulton Differential Revision: D75886336 fbshipit-source-id: 9a55cb7b1ccde48783e387be4e8c9da8b56a9c08
Summary: Pull Request resolved: #3869 This commit adds a new color_constants.py file to that centralizes color definitions for Ax visualizations. The file imports base colors creates semantic aliases that clearly describe their usage contexts. The primary goal is to encourage consistent color usage across the Ax visualization system by: 1. Providing descriptive, semantic names for colors based on their functional purpose 2. Creating a single source of truth for color definitions 3. Making color schemes more discoverable for developers Reviewed By: eonofrey Differential Revision: D75886310 fbshipit-source-id: 719708ddc163d96c527aa50385befa96dfe91714
Summary: Pull Request resolved: #3861 Adding auxiliary experiment methods to Experiment class to add (as first experiment) and find auxiliary experiment. Meta: Add to first is needed as we currently support one aux PE experiment and the first one is the one with highest priority and is the one that will be used. Find aux experiment is needed to support pref opt config (see the diff on top), which will allow us to identify which PE exp to use instead of relying on the order. Reviewed By: mgarrard Differential Revision: D75696082 fbshipit-source-id: cff96ee75ba29f5831e3136ee47df0d7d931f9b3
Summary: Pull Request resolved: #3844 This diff updates the current `gen` method to be called `gen_single_trial` -- we will update gen_multiple_trials_with_multiple_models to be called gen in the next diff (initially tried to do them in the same diff and it was very confusing). I decided to replace everything that currently calls gen with gen_single_trial instead of replace some of them with gen_multi as i think it kept things cleaner. This primarily effects test files. Reviewed By: lena-kashtelyan, esantorella Differential Revision: D75688420 fbshipit-source-id: 2b338f56fcd56e237394b52a609d92e1e2293b7a
Summary: Pull Request resolved: #3846 We want the most powerful gen to be our default gen, and that's gen_for_multi_with_multi, in the previous diff we free'd up the name gen Reviewed By: saitcakmak Differential Revision: D75718755 fbshipit-source-id: 4670e5f68d7828ca32d965ee614f01ad274c3726
Summary: Pull Request resolved: #3856 We will switch to using `ExperimentData` rather than `list[Observation]` within the `Adapter`, which necessitates replacing `list[Observation]` input to `Transform` constructors with `ExperimentData`, and supporting transforms with `ExperimentData`. This diff only updates the `Transform` constructors to accept the new argument and adds `Transform.transform_experiment_data` to the base class. The transforms will be updated in following diffs to make use of the new argument in the constructor and implement the `transform_experiment_data` method. Background: As part of the larger refactor, we will be using `ExperimentData` in place of `list[Observation]` within the `Adapter`. - The transforms will be initialized using `ExperimentData`. The `observations` input to the constructors may be deprecated once the use cases are updated. - The training data for `Adapter` will be represented with `ExperimentData` and will be transformed using `transform_experiment_data`. - For misc input / output to various `Adapter` and other methods, the `Observation / ObservationFeatures / ObservationData` objects will remain. To support these, we will retain the existing transform methods that service these objects. - Since `ExperimentData` is not planned to be used as an output of user facing methods, we do not need to untransform it. We are not planning to implement`untransform_experiment_data`. Reviewed By: esantorella Differential Revision: D72664137 fbshipit-source-id: 27843bcaa751676f5684a7d0b8c52e4cbe46a75e
Summary: Pull Request resolved: #3871 As titled. Supports transforming `ExperimentData` with `IntToFloat` transform. Background: As part of the larger refactor, we will be using `ExperimentData` in place of `list[Observation]` within the `Adapter`. - The transforms will be initialized using `ExperimentData`. The `observations` input to the constructors may be deprecated once the use cases are updated. - The training data for `Adapter` will be represented with `ExperimentData` and will be transformed using `transform_experiment_data`. - For misc input / output to various `Adapter` and other methods, the `Observation / ObservationFeatures / ObservationData` objects will remain. To support these, we will retain the existing transform methods that service these objects. - Since `ExperimentData` is not planned to be used as an output of user facing methods, we do not need to untransform it. We are not planning to implement`untransform_experiment_data`. Reviewed By: esantorella Differential Revision: D75903694 fbshipit-source-id: 8cad41c0734a7456ff4539a3b824df9100f8f578
Summary: Pull Request resolved: #3872 As titled. Supports transforming `ExperimentData` with `Log` transform. Background: As part of the larger refactor, we will be using `ExperimentData` in place of `list[Observation]` within the `Adapter`. - The transforms will be initialized using `ExperimentData`. The `observations` input to the constructors may be deprecated once the use cases are updated. - The training data for `Adapter` will be represented with `ExperimentData` and will be transformed using `transform_experiment_data`. - For misc input / output to various `Adapter` and other methods, the `Observation / ObservationFeatures / ObservationData` objects will remain. To support these, we will retain the existing transform methods that service these objects. - Since `ExperimentData` is not planned to be used as an output of user facing methods, we do not need to untransform it. We are not planning to implement`untransform_experiment_data`. Reviewed By: SebastianAment Differential Revision: D75905207 fbshipit-source-id: 5950f64a6b2757f8fd5173433585ea198d4fc232
Summary: Pull Request resolved: #3877 Updates naming of color constants used throughout the Ax codebase Reviewed By: eonofrey Differential Revision: D76043167 fbshipit-source-id: aaed8c346aeb2703c8b66f5355a91b520f9e7078
Summary: Pull Request resolved: #3857 The `no_bayesian_optimization` arg has been replaced by `force_random_search`, and deprecation messages have been in place for 9 months (#2693). This diff deprecates. Reviewed By: ltiao, SebastianAment, saitcakmak Differential Revision: D75891247 fbshipit-source-id: e75464390ba24e8f0d85afe423af1c9df944fa30
Summary: Pull Request resolved: #3884 included is a minor fix that checks status of status_quo and doesn't draw the line if it is a candidate. It does that by checking if the metric mean is calculated or not. Here is a working demo: https://fburl.com/anp/84ecgfep Reviewed By: mpolson64 Differential Revision: D75976043 fbshipit-source-id: ad75c05a6dbdbeb5b20e673ab3b67026798b74fb
Summary: Pull Request resolved: #3873 As titled. Supports transforming `ExperimentData` with `RemoveFixed` transform. Background: As part of the larger refactor, we will be using `ExperimentData` in place of `list[Observation]` within the `Adapter`. - The transforms will be initialized using `ExperimentData`. The `observations` input to the constructors may be deprecated once the use cases are updated. - The training data for `Adapter` will be represented with `ExperimentData` and will be transformed using `transform_experiment_data`. - For misc input / output to various `Adapter` and other methods, the `Observation / ObservationFeatures / ObservationData` objects will remain. To support these, we will retain the existing transform methods that service these objects. - Since `ExperimentData` is not planned to be used as an output of user facing methods, we do not need to untransform it. We are not planning to implement`untransform_experiment_data`. Reviewed By: esantorella Differential Revision: D75906004 fbshipit-source-id: 39403b3c791bc7514a4e6771a3d773e41006857a
Summary: Pull Request resolved: #3874 As titled. Supports transforming `ExperimentData` with `ChoiceToNumericChoice` transform and subclasses. Background: As part of the larger refactor, we will be using `ExperimentData` in place of `list[Observation]` within the `Adapter`. - The transforms will be initialized using `ExperimentData`. The `observations` input to the constructors may be deprecated once the use cases are updated. - The training data for `Adapter` will be represented with `ExperimentData` and will be transformed using `transform_experiment_data`. - For misc input / output to various `Adapter` and other methods, the `Observation / ObservationFeatures / ObservationData` objects will remain. To support these, we will retain the existing transform methods that service these objects. - Since `ExperimentData` is not planned to be used as an output of user facing methods, we do not need to untransform it. We are not planning to implement`untransform_experiment_data`. Reviewed By: esantorella Differential Revision: D75967458 fbshipit-source-id: 9f70c35f85b80761df2a87adbf1db433cbb3fe7b
Summary: Pull Request resolved: #3875 As titled. Supports transforming `ExperimentData` with `FillMissingParameters` transform. Background: As part of the larger refactor, we will be using `ExperimentData` in place of `list[Observation]` within the `Adapter`. - The transforms will be initialized using `ExperimentData`. The `observations` input to the constructors may be deprecated once the use cases are updated. - The training data for `Adapter` will be represented with `ExperimentData` and will be transformed using `transform_experiment_data`. - For misc input / output to various `Adapter` and other methods, the `Observation / ObservationFeatures / ObservationData` objects will remain. To support these, we will retain the existing transform methods that service these objects. - Since `ExperimentData` is not planned to be used as an output of user facing methods, we do not need to untransform it. We are not planning to implement`untransform_experiment_data`. Reviewed By: esantorella Differential Revision: D75972103 fbshipit-source-id: 739f332fc2e8d7ae1f749ff7e3562ba0e1a9ce7a
Summary: Pull Request resolved: #3876 As titled. Supports transforming `ExperimentData` with `Logit` transform. Background: As part of the larger refactor, we will be using `ExperimentData` in place of `list[Observation]` within the `Adapter`. - The transforms will be initialized using `ExperimentData`. The `observations` input to the constructors may be deprecated once the use cases are updated. - The training data for `Adapter` will be represented with `ExperimentData` and will be transformed using `transform_experiment_data`. - For misc input / output to various `Adapter` and other methods, the `Observation / ObservationFeatures / ObservationData` objects will remain. To support these, we will retain the existing transform methods that service these objects. - Since `ExperimentData` is not planned to be used as an output of user facing methods, we do not need to untransform it. We are not planning to implement`untransform_experiment_data`. Reviewed By: esantorella Differential Revision: D75973054 fbshipit-source-id: a23f9c27f396e5328f0f83293ec9454c894a7788
Summary: Pull Request resolved: #3886 As titled. Supports transforming `ExperimentData` with `OneHot` transform. Background: As part of the larger refactor, we will be using `ExperimentData` in place of `list[Observation]` within the `Adapter`. - The transforms will be initialized using `ExperimentData`. The `observations` input to the constructors may be deprecated once the use cases are updated. - The training data for `Adapter` will be represented with `ExperimentData` and will be transformed using `transform_experiment_data`. - For misc input / output to various `Adapter` and other methods, the `Observation / ObservationFeatures / ObservationData` objects will remain. To support these, we will retain the existing transform methods that service these objects. - Since `ExperimentData` is not planned to be used as an output of user facing methods, we do not need to untransform it. We are not planning to implement`untransform_experiment_data`. Reviewed By: esantorella Differential Revision: D76074577 fbshipit-source-id: 07f17b5f91a7e1ce5c6d9eb91dd034bec17d6f89
Summary: Pull Request resolved: #3891 * Speed up sensitivity computations by using batch dimension -- use an x of shape [n, 1, d] instead of [n, d]. This avoids the costly creation of an (n, n) covariance matrix when we only need its diagonal elements. * Stop doing those computations in minibatches, which was done to avoid the superlinear memory usage from large batch sizes (Use mini batches in SobolSensitivityGPMean #1848 ), which came from the (n,n) covariance matrix we no longer compute. Reviewed By: Balandat Differential Revision: D75712208 fbshipit-source-id: 19714227a6f0124064b3f4576ab8bd34eca3c708
Summary: Pull Request resolved: #3892 Two arguments are no longer used, one as a result of D75712208 and one has long been unused Reviewed By: saitcakmak Differential Revision: D76151380 fbshipit-source-id: f9572bb76282b90001c1c50bbe173900bc8e4b11
Summary: Pull Request resolved: #3887 As titled. Supports transforming `ExperimentData` with `BilogY` transform. Also removes unnecessary `DataRequiredError`, since the transform does not actually utilize the provided data in the constructor. Background: As part of the larger refactor, we will be using `ExperimentData` in place of `list[Observation]` within the `Adapter`. - The transforms will be initialized using `ExperimentData`. The `observations` input to the constructors may be deprecated once the use cases are updated. - The training data for `Adapter` will be represented with `ExperimentData` and will be transformed using `transform_experiment_data`. - For misc input / output to various `Adapter` and other methods, the `Observation / ObservationFeatures / ObservationData` objects will remain. To support these, we will retain the existing transform methods that service these objects. - Since `ExperimentData` is not planned to be used as an output of user facing methods, we do not need to untransform it. We are not planning to implement`untransform_experiment_data`. Reviewed By: esantorella Differential Revision: D76081492 fbshipit-source-id: e4c59f64ffc9a9e34701be9db54ad2b9e81569ef
Summary: Pull Request resolved: #3888 As titled. Supports transforming `ExperimentData` with `UnitX` transform. Background: As part of the larger refactor, we will be using `ExperimentData` in place of `list[Observation]` within the `Adapter`. - The transforms will be initialized using `ExperimentData`. The `observations` input to the constructors may be deprecated once the use cases are updated. - The training data for `Adapter` will be represented with `ExperimentData` and will be transformed using `transform_experiment_data`. - For misc input / output to various `Adapter` and other methods, the `Observation / ObservationFeatures / ObservationData` objects will remain. To support these, we will retain the existing transform methods that service these objects. - Since `ExperimentData` is not planned to be used as an output of user facing methods, we do not need to untransform it. We are not planning to implement`untransform_experiment_data`. Reviewed By: esantorella Differential Revision: D76085733 fbshipit-source-id: 40a07675cfc4252f9d5699423e6e2bfea5295378
Summary: Pull Request resolved: #3893 These tests were removed in D74846187 because they were slow. D75569748 and D75712208 sped them up. Reviewed By: saitcakmak Differential Revision: D75712258 fbshipit-source-id: 24dd33697d9db4ec6bcc9ebf4a727181a8873906
Summary: Pull Request resolved: #3885 Arm order in metric results were displayed per the data order. This led to a random ordering of arms in the UI {F1977891605} To fix this, we want to sort the arm ordering in data. Data will be sorted by trail_index, then arm_name. Arm name will be sorted as 'custom_name' < '0_1' < '0_2' < '0_11' < '0_100' Reviewed By: lena-kashtelyan, saitcakmak Differential Revision: D74409276 fbshipit-source-id: 51d4de4e50f20ac009ed29ec58d4a417be2857a1
#4115) Summary: Pull Request resolved: #4115 If any progression value is missing in `ObservationFeatures.metadata`, `MapKeyToFloat._transform_observation_feature` fills it with the upper bound of the corresponding parameter. This encodes the assumption that any data without a progression value corresponds to a completed trial (as a trial can be completed with `client.complete_trial` without providing progression). The initial implementation of `transform_experiment_data` did not account for the possibility of NaN values in the corresponding index. This diff updates the method to check for NaN progression values and fill them with the target progression. Reviewed By: ltiao Differential Revision: D79750521 fbshipit-source-id: f5aef946fe55db84af39edb10fd7d9499ed2ada3
…iginal (#4116) Summary: Pull Request resolved: #4116 This transform was popping the corresponding element from the metadata before adding it as a feature. This is fine for the most part, but it doesn't really have a significant benefit over just leaving the original in the metadata. The problem with popping the element from the metadata is when doing this with `ExperimentData`. Even if we `deepcopy` a pandas dataframe, it does not deepcopy the underlying dictionary values. If we then transform the copy, we end up removing the metadata from the original as well, which leads to errors when doing cross validation with an adapter that uses these transforms (since we end up transforming multiple copies, and only the first one has the metadata). This diff updates the transfrom to not remove the element from metadata. Reviewed By: ltiao Differential Revision: D79805146 fbshipit-source-id: 75ee5da5425b4fbb750c0041fbe58f8a5e4e5a44
Summary: Pull Request resolved: #4117 Reviewed By: mpolson64 Differential Revision: D79741086 fbshipit-source-id: 2b75bb01b7b7eaba0c3478f60d20e97c050f5e59
Summary: Pull Request resolved: #4112 This diff groups the cross-validation plots in the `DiagnosticAnalysis` class of Ax. Simplify by removing unnecessary unpacking and `.flatten()` call. Reviewed By: mpolson64 Differential Revision: D79558715 fbshipit-source-id: a833c9c31c8a90fa9faf3de9e61ba426d15b79df
Summary: Pull Request resolved: #4118 ### Added Filtering by Trial Status to Summarize This diff adds a new feature to the `summarize` method, allowing users to filter by trial status. #### What's Changed Reviewed By: mpolson64 Differential Revision: D79442982 fbshipit-source-id: 31a9bfc6b7ccc9c925dc09870a5346e4dea49427
Summary: Pull Request resolved: #4119 This diff updates the docstrings in `client.py` and `summary.py` to improve code quality, making it easier for users to understand the purpose and usage of the functions. Specifically, the changes include: * Adding function argument explanations for `trial_indices` and `trial_status` in the `client.py` file. * Adding function argument explanations for `trial_indices`, `trial_status`, and `omit_empty_columns` in the `summary.py` file. These updates enhance the overall clarity and readability of the codebase, ensuring that users can effectively utilize the functions and their arguments. Reviewed By: mpolson64 Differential Revision: D79738530 fbshipit-source-id: 54b3f379399105eec0b4db4f070625280ea71fd7
Summary: Pull Request resolved: #4101 ModelConfig has a name field that is optional. MBM Surrogate needs an identifier for model; currently uses name if present, or just drops the model if not when storing eval results. MBM Generator needs an identifier for model and uses name if present, or str(model_config) if not. This cleans things up a bit by providing an identifier for ModelConfig that will always exist and so can be used in the place of name and clean up logic in the other parts of MBM. The identifier follows what was done in MBM Generator for tracking gen metadata: it is name if name exists, otherwise a string dump of the model config. Reviewed By: sdaulton Differential Revision: D78560884 fbshipit-source-id: 89a94dbeaf97d799e46a7d12f248cdf4382d3ea2
Summary: Pull Request resolved: #4109 Adds a "models_for_gen(n)" method to Surrogate that will constructs a list of n models from which to generate n candidates. The list of models is constructed by sampling from the given ModelConfigs according to their model selection eval criterion (taken as a measure of model quality). Reviewed By: sdaulton Differential Revision: D77252555 fbshipit-source-id: 89c82ce62b47ee13569fe42cdff5427b4bdddf30
Summary: Pull Request resolved: #4110 Augments MultiAcquisition to additionally handle the case where we want to generate from multiple models with the same botorch acquisition class. The collection of models from which to generate is created by Surrogate. This requires making the MBM Acquisition aware of how many points are intended to be generated with this particular Acquisition. Note that we do construct a new Acquisition in every call to gen. The diff does a refactor of `Acquisition.__init__` to move some of the logic into other methods that are called during `__init__`, and then can be re-called when we want to use the same Acquisition object to instantiate multiple botorch acquisition functions with different models each. The `gen_metadata` entry that stores which model was used for the generation is now potentially a list of models, so I renamed the entry, both to be something a little more clear and also because there is a breaking change in the data type if one were trying to do some type of meta analysis anyway. Reviewed By: sdaulton Differential Revision: D78560935 fbshipit-source-id: 5bfa915d089df7ff9999fd7b3c264d5944f0c09e
Summary: Pull Request resolved: #4120 As titled, update color sheme for ContourPlot and Sensitivity Analysis to be Earth themed over current PrGn Reviewed By: mgarrard Differential Revision: D79893711 fbshipit-source-id: 874073cf2221f398561b754e4872fe8c64b30c80
Summary: Pull Request resolved: #4044 In line with the logging present in AxClient. Adds logging on: * [INFO] Result of choose_gs * [INFO] Candidate generation * [INFO] Trial status marking * [WARNING] if user calls complete_trial but trial is marked failed due to missing metrics * [INFO] Early stopping reason Reviewed By: saitcakmak Differential Revision: D78183140 fbshipit-source-id: 41d4470b5debeb436e2f2c5adc5720a6138df645
Summary: Pull Request resolved: #4122 Reviewed By: saitcakmak Differential Revision: D79907818 fbshipit-source-id: 572b461f3df8171da2d1147f222b733a5988d5b6
Summary: Pull Request resolved: #4123 It is possible to fit a TorchAdapter with only one data point but it is not possible to run cross validation with it. This can lead to errors since we rely on CV statistics in `BestModelSelector`, which is used in BOTL. This diff gracefully handles such errors rather than erroring out in the experiment. Reviewed By: esantorella Differential Revision: D79915007 fbshipit-source-id: 0dd8a1f5359cee5a12035fd5321c4a28dba2107a
Summary: Pull Request resolved: #4124 See #4103 for motivation. We need to check for search space membership when we sample from the surrogate to draw the contour plot, specifically because in experiments where the status quo arm is all None we were getting errors. However, we dont actually care if points are out of sample is technically out of the search space due to parameter constraint violation (ex. if the value for a non-plotted parameter selected by select_fixed_value causes some parameter constraint to be violated we shouldnt mind and Ax should go ahead and plot regardless). By moving the check higher up we avoid this error entirely. Reviewed By: mgarrard Differential Revision: D79917972 fbshipit-source-id: 840d83b8c97969e22cbd8f5de76705360eb7419c
Summary: Pull Request resolved: #4126 ### Context In the benchmarking set-up, the `normalize_progression=True` option in early-stopping strategies is currently not working properly as zero trials are early-stopped. Specifically, it appears that `is_eligible_any` returns `False` for all trials. All early-stopping strategies (all subclasses of `BaseEarlyStoppingStrategy`) are impacted. Specifically the instantiation of `MapData` in the return call of `_check_validity_and_get_data` casts the progression column to `int64`, which rounds all the float-valued progressions `[0.0, 1.0)` to `0`, leading `is_eligible_any` to always return `False` ### Changes This diff removes the specification of `map_key_info` as integer-typed in `BenchmarkMapMetric` and `BenchmarkMapUnavailableWhileRunningMetric`. Reviewed By: esantorella Differential Revision: D79938692 fbshipit-source-id: 6e283d8050606c292d2572994aa8925af8c6a477
…4127) Summary: Pull Request resolved: #4127 With fully bayesian models, this was leading to quite a large memory usage with the default batch size of 2048. With this change, the max batch size will adapt to the model batch shape, which should bring down the memory usage of batched models to a level similar to non-batched models. Reviewed By: dme65 Differential Revision: D80006058 fbshipit-source-id: 39e17b37678890822afd36f36aad3e72a6bd6c42
Summary: Pull Request resolved: #4129 With logging changes from D78183140 these print statements are extraneous, see screenshot. This tidies things up quite a bit. {F1981106230} Reviewed By: saitcakmak Differential Revision: D80030926 fbshipit-source-id: d4c4f198969224c13fe60d712ac0950fe47cb818
Summary: Pull Request resolved: #4125 Reviewed By: CristianLara Differential Revision: D80031830 Pulled By: Balandat fbshipit-source-id: 080f1b8f1a13944062a1299b4f4cea266cae92e2
Summary: Pull Request resolved: #4130 As titled. Also bumped the pinned botorch version to 0.15.0 Notable items on the changelog come from a skim of 1.0.0...main In the future we should make a point to release more often (1-2 times a month) to avoid having such a large swath of commits to skim through (in this case we had 321 changes since Ax 1.0.0) Reviewed By: saitcakmak Differential Revision: D80030270 fbshipit-source-id: 177088348a8a653a66b64d9d026ee85bccb8db1e
Summary: Pull Request resolved: #4133 Fix bug where `ArmEffectsPlot` and `ScatterPlot` are showing the latest CANDIDATE Trial twice. This is currently happening because the following line: ``` ~df["trial_status"].isin([ts.name for ts in FAILED_ABANDONED_STATUSES]) ``` brings in candidates before we additionally append `candidate_trial` a few lines later. By including candidate trials in the exclusion filter (i.e. `FAILED_ABANDONED_CANDIDATE_STATUSES`), we avoid duplicate candidate trials Reviewed By: Cesar-Cardoso Differential Revision: D80046532 fbshipit-source-id: 947b64ac6695a0f51308fd376a43e3cb3087c01a
Summary: Pull Request resolved: #4135 We need to repin botorch in Ax since D78560935 landed in between when we released botorch 0.15.0 and when we tried to deploy Ax 1.1 Reviewed By: esantorella Differential Revision: D80044408 fbshipit-source-id: 10cc55b9c9b026e2ab3652214be7bf79e86ca1a9
meta-cla
bot
added
the
CLA Signed
Contributor
Author
|
I double checked that the merge from main into this branch only included commits up to and including the new release tag 0c0c892 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
20 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
back reference
Ax counterpart to the same PR in botorch meta-pytorch/botorch#2969
context
The release workflow failed while creating the new docusaurus version for 1.1.0 https://github.com/facebook/Ax/actions/runs/16917549143/job/47935175650
solution
docusaurus-versionsas it's not a protected branch but figured it'd be good to make this a PRThe result is essentially two commits: a merge commit to catch up with master, and the docusaurus versioning commit
Once this PR is merged we can trigger the Publish Website workflow from the UI (or let the nightly job run it) which will deploy these latest website changes
notes
Note that this branch is being merged not into
main, but into thedocusaurus-versionsrelease branch. We don't commit docusaurus version to main as the "Meta CLA Check" does not permit our Github Actions workflow to commit tomain.test plan
Here's the result of running
./scripts/make_docs.shon this branch: (version picker starts to fade while I click for screenshot...)