Snowflake-ml-python

Behavior Changes

- Use cloudpickle to serialize and deserialize models throughout the codebase and removed dependency on joblib.

New Features

- Model Deployment: Added support for snowflake.ml models.

Behavior Changes

- Standardized registry API with following
- Create & open registry taking same set of arguments
- Create & Open can choose schema to use
- Set_tag, set_metric, etc now explicitly calls out arg name as metric_name, tag_name, metric_name, etc.

New Features

- Changes to support python 3.9, 3.10
- Added kBinsDiscretizer
- Support for deployment of XGBoost models & int8 types of data

Behavior Changes

- Big Model Registry Refresh
- Fixed API discrepancies between register_model & log_model.
- Model can be referred by Name + Version (no opaque internal id is required)

New Features

- Model Registry: Added support save/load/deploy SKL & XGB Models

Bug Fixes

- Allow using OneHotEncoder along with sklearn style estimators in a pipeline.

New Features

- Model Registry: Added support for delete_model. Use delete_artifact = False to not delete the underlying model data
but just unregister.

New Features

- Initial version of snowflake-ml modeling package.
- Provide support for training most of scikit-learn and xgboost estimators and transformers.

Bug Fixes

- Minor fixes in preprocessing package.

New Features

- New in Preprocessing:
- SimpleImputer
- Covariance Matrix
- Optimization of Ordinal Encoder client computations.

Bug Fixes

- Minor fixes in OneHotEncoder.