Skip to content

CHANGELOG.md

Latest commit

 

History

History
39 lines (28 loc) · 2.35 KB

CHANGELOG.md

File metadata and controls

39 lines (28 loc) · 2.35 KB

spacebench (developing)

Added

  • Use new covariate groups and confounding scores per metric. Confounding scores are now a dictionary with keys ate, erf, ite, and importance. The first three correspond to the error scores (absolute ate error, mean absolute erf error, mean root-mean-squared ite error) using a baseline model from autogluon. The importance scores correspond to the max of min(outcome_importance, treatment_importance), which are the previously used scores.
  • Updated env.py to allow for different data formats since data collections in space-data can now use .parquets, .graphmlz and other compressed formats to make read speed and storage more efficient.
  • Updated env.py to read index col as string for tab and csv files to avoid issues with integer indices.

Changed

  • Fixed bug in Space Dataset where smoothness_score was called `snoothness scores``.
  • confounding_score and smoothness_score are not both singular in a SpaceDataset.
  • Mask entire covariate groups from new covariate groupping in the space environments.

Removed

  • Remove option to filter by confounding score in make. It is not useful since user can filter externally examining the confounding_scores attribute.

v0.0.2

Added

  • DAPSm algorithm. Basic functionality with optimal and greedy matching.
  • Evaluator classes with demo on evaluator_demo.ipynb
  • SpaceEnv classes (same demo). Data automatically downloads given name.
  • Added Masterfile but needs to be updated with new datasets.
  • list_envs in DataMaster (previously list_datasets) now has an option for binary=True or continuous=True
  • Added examples of benchmarks in the examples/ folder.
  • Add nice printing methods for SpaceDataset and SpaceEnv.
  • Warning messages about data limitations are now printed when loading the package, creating a SpaceEnv, or creating a SpaceDataset.

Changed

  • Deleted spacebench/datasets/ folder in favor of spaceenv/env.py more consie module.
  • CausalDataset --> SpaceDataset
  • SpaceEnv, SpaceDataset DataMaster are imported directly from spacebench in spacebench/__init__.py.
  • Datamaster.list_datasets -> Datamaster.list_envs
  • pip install will now install all dependencies automatically. It will not install dependencies of specific algorithms or examples.
  • New optional [all] in pip install for the dependencies used in examples and algorithms.