Tuning hyperparameters is often the most expensive part of the machine learning workflow. Tune is built to address this, demonstrating an efficient and scalable solution for this pain point.
Mailing List https://groups.google.com/forum/#!forum/ray-dev
- exercise_1_basics.ipynb covers basics of using Tune - creating your first training function and using Tune. This tutorial uses Keras.
- exercise_2_optimize.ipynb covers Search algorithms and Trial Schedulers. This tutorial uses PyTorch.
Concepts that are generally useful but have not been covered:
- Using PBT
- Creating a Trainable with save and restore functions and checkpointing
- Distributed execution on a larger cluster
Please open an issue if you have any questions or identify any issues. All suggestions and contributions welcome!