Tutorial 4
Deep Learning Workflows with Skorch and RAPIDS
This notebook focuses on model development workflows that combine Skorch with GPU-enabled tooling, giving attendees a reusable pattern for deep learning experimentation in notebook form.
Distributed Hyperparameter Optimization with Ray Tune and RAPIDS
What is RayTune?
RayTune is a scalable Hyperparameter optimization library. It allows distributed HPO, provides various search algorithms to allow different optimization techniques to be explored with ease. The library also provides scheduling algorithms that allows a smarter way to schedule the different parameter sweep instead of the basic First In-First Out method which is followed by other libraries (Scikit-Learn, Dask-ml) that support HPO. The different scheduling algorithms can make the HPO process resource efficient and help arrive at the best parameters much faster.
What you will do
- open a ready-to-run Colab notebook
- work through a GPU-based deep learning training workflow
- experiment with tuning and evaluation in an interactive environment
- adapt the notebook for follow-up research projects