Noah Brenowitz, UW eScience & Atmospheric Sciences
In this talk, I will discuss the tools behind some of my recent work building machine-learned components for climate models. A typical machine learning pipeline consists of three stages: 1) data munging , 2) model training, and 3) model evaluation. The corresponding libraries I use for each of these stages are 1) xarray, 2) scikit-learn and PyTorch, and 3) HoloViews and matplotlib for visualization. I will demonstrate how all these tools come together on a simple example problem. Time permitting, I will also discuss how to embed python-based machine learning models within a Fortran climate model.
Go to the presentation materials, including a link to Binder where you can run the Jupyter notebooks