v0.5.0

Breaking Changes

• Minimum Python version raised to 3.9 (dropped support for 3.7, 3.8)
• Minimum PySpark version raised to 3.4 (dropped support for 3.2, 3.3)

New Features

• Added support for Python 3.11, 3.12, 3.13

Bug Fixes

• Added hnswlib as fallback for nmslib on macOS ARM (fixes segfault in metric split)

Dependencies

• Updated numpy to >=1.24.0, <3.0.0
• Updated pandas to >=1.5.0, <3.0.0
• Updated scipy to >=1.10.0
• Updated scikit-learn to >=1.3.0
• Updated nmslib to >=2.1.0
• Added hnswlib >=0.7.0 as alternative KNN backend
• Updated catboost to >=1.2.0

Internal

• Replaced deprecated pkg_resources with importlib.metadata
• Updated CI/CD to test Python 3.9-3.13

Full Changelog: https://github.com/MobileTeleSystems/Ambrosia/blob/main/CHANGELOG.rst

Ambrosia 0.4.1

Hotfix for pyspark import in spark criteria.

Ambrosia 0.4.0

• Documentation and usage examples have been substantially reworked and updated.

• The Designer class and design methods functionality is updated.

  • Empirical design now supports the choice of hypothesis alternative and group ratio parameter

  • Look of resulting tables with calculated parameters is unified for all design methods

  • Changed multiprocessing strategy for bootstrap criterion

• The Tester class functionality is updated.

  • Spark data support for the Tester class is added. Independent t-test is available now

  • Bootstrap criterion can now return deterministic output using a random_seed parameter

  • Paired bootstrap criterion is now available

  • MHC now is optional and takes into account the number of passed metrics

  • first_errors parameter renamed to first_type_errors

• pyspark package now is optional and could be installed using pip extras.

• Fixed a set of bugs.

Ambrosia 0.3.0

In these release we introduce the following updates:

• The Designer theoretical methods now can be used for the binary data
• The Designer theoretical methods now supports hypothesis alternative and group ratio parameters
• All individual data processing classes have been updated to use the fit and transform methods
• IQRPreprocessor, BoxCoxTransformer, LogTransformer classes have been added in ambrosia.preprocessing
• The Preprocessor class now can store all transformation in one json file
• The MLVarianceReducer can store and load picklable ML model
• Other changes in data processing classes

You can see the detailed changelog here: https://github.com/MobileTeleSystems/Ambrosia/blob/main/CHANGELOG.rst

Ambrosia 0.2.0

Library name changed back to ambrosia. Naming conflict in PyPI has been resolved.
0.1.x versions are still available in PyPI under ambrozia name.

Ambrosia 0.1.2

T-test absolute effect calculation bug fix.

Ambrosia 0.1.1

Hotfix for library naming.
Library temporary renamed to ambrozia in PyPI repository due to hidden naming conflict.

Ambrosia 0.1.0

First release

First release of Ambrosia package:

• Added Designer class for experiment parameters design
• Added Spliiter class for A/B groups split
• Added Tester class for experiment effect measurement
• Added various classes for experiment data preprocessing
• Added A/B testing tools with wide functionality