Collie

Fixed
- Versions of PyTorch Lightning between ``1.6.0`` and ``1.9.X`` are now supported

Added
- ``HybridModel`` and ``HybridPretrainedModel`` now take additional optional parameters ``user_metadata`` and ``user_metadata_layers_dims``
- ``get_data.py`` now includes ``get_user_metadata``
Changed
- Added ``item_metadata_layers_dims`` and ``user_metadata_layers_dims`` parameters to ``HybridPretrainedModel`` and ``HybridModel`` and removed ``metadata_layers_dims``
- Updated notebooks and examples to include usage of ``user_metadata``

Fixed
- a ``Value Error`` is now raised when ``item_metadata`` contains nulls

Fixed
- Deprecated PyTorch Lightning unit test

Added
- option to ``force_split`` to ``stratified_split``
- better type hints for ``Callable``s
- added methods ``get_user_predictions`` and ``user_user_similarity`` to the ``BasePipeline``
- added ``_get_user_embeddings`` method to all model classes
Changed
- default ``Dockerfile`` image to be ``torch1.10.0`` with CUDA 11.3
- check if index is in-bound for ``get_item_predictions`` and ``item_item_similarity`` before calling the model
- added ``enable_model_summary`` and ``detect_anomaly`` parameters to ``CollieMinimalTrainer`` and deprecated ``weights_summary`` and ``terminate_on_nan`` to more closely match the new ``pytorch_lightning`` API
- clarified error message when user has a single interaction when using ``stratified_split``
- updated all examples, tests, and notebooks with post-1.5.0 PyTorch Lightning APIs
Fixed
- device error when running metrics for a ``MultiStagePipeline`` models
- ``CollieMinimalTrainer`` model summary to work with later versions of PyTorch Lightning
Removed
- default ``num_workers`` for ``Interactions`` DataLoaders

Added
- string support for Adagrad optimizer in model pipelines