Ml4ir

Changed
- Upgraded min version for project from python37 to python39
- Dependency versions in requirements.txt

Added
- RankMatchFailure metric for evaluation
- Statistical significance and power analysis utilities
- Stat analysis for groupwise metrics in Ranking

Changed

- Upgrading from tensorflow 2.0.x to 2.9.x
- Moving from Keras Functional API to Model Subclassing API for more customization capabilities
- Auxiliary loss is reimplemented as part of ScoringModel

Added

- AutoDAGNetwork which allows for building flexible connected architectures using config files
- SetRankEncoder keras Layer to train SetRank like Ranking models
- Support for using tf-models-official deep learning garden library
- RankMatchFailure metric for validation

Changed

- Ability to pass custom RelevanceModel class in Pipeline.

Fixed

- Bug in metrics_helper when used without secondary_labels

Added

- RankMatchFailure metric for evaluation
- RankMatchFailure auxiliary loss

Changed

- Adding rank feature to serving parse fn by default and removing dependence on required serving_info attribute