- Added support for image-based models, including new image-based feature transformation operations and TensorFlow and PyTorch data loading functions for image data.
- Added support for user-defined feature transformation operations, specifically including the ability to define custom training data feature transformation operations in python, and upload them for usage in the Lucd GUI.
- Added functions to eda.lib.lucd_ml module for simplifying the process of getting classification and regression predictions from TensorFlow estimators, and well as computing confusion matrices.
- Updated modeling framework to be compatible with TensorFlow v2.1 (from v1.3).
- Removed the “simplified/templated modeling approach”. Now users provide models via full python scripts.
- Fixed bugs in loading data into models from the Lucd unified dataspace.
- Fixed bugs in loading data for multiclass (> 2 classes) TensorFlow text classification models.