Skip to content

Latest commit

 

History

History
69 lines (57 loc) · 4.46 KB

README.md

File metadata and controls

69 lines (57 loc) · 4.46 KB

Model Evaluation (2023)

Training Overview

This HyTEST model evaluation course teaches students the foundations of scientific model evaluation, as well as how to apply those foundations to evaluate big datasets in the cloud using Pangeo, an open-source Python platform for distributed computing.

Using a combination of lectures and hands-on exercises in Jupyter Notebooks, we will cover data access and preparation, identifying and computing statistical evaluation metrics, and visualization of results. Furthermore, we consider how workflows can be modified for individual purposes and/or other use-cases.

Learning Outcomes

On the technical side, students will become familar with fundamental components of the Pangeo architecture, including S3, Dask, intake, xarray, and hvplot. On the theory side, students will learn about theory of model evaluation, how to compute and interpret different metrics for benchmarking model performance. By the end, the student should be able to take the output of a geoscientific model, choose a meaningful suite of performance metrics, then use Pangeo to compute them and visualize the result.

Date/Time

May 10th and May 11th of 2023, 10am to 2pm MT

Lesson plan

Day 1

  • Welcome/Icebreaker - 20 minutes
  • Lesson 0: HyTEST, Pangeo Concepts, Tools - 20 minutes
  • Lesson 1: Training Setup - 20 minutes
  • Break - 15 minutes
  • Lab 1: Data exploration - 30 minutes
  • Break - 10 minutes
  • Lab 2: Hvplot demonstration - 35 minutes
  • Q & A: Day 1 - 20 minutes

Day 2

  • Lesson 3: Standard benchmarks (theory) - 30 minutes
  • Lab 3: Standard benchmarks (application) - 30 minutes
  • Break - 10 minutes
  • Lesson 4: Selecting your objective - 30 minutes
  • Break - 30 minutes
  • Lab 4: Selecting your objective - 30 minutes
  • Lesson 5: d-score benchmark (theory) - 30 minutes
  • Break - 10 minutes
  • Lab 5: d-score benchmark (application) - 20 minutes
  • Wrap-up and Q&A - 20 minutes

Presenters

Tim Hodson (USGS), Erin Towler (NCAR), Gene Trantham (USGS), & Sydney Foks (USGS)

Course Preparation

Prior knowledge of Jupyter Notebooks and Python will be beneficial to participants, but not required. Please review some of the links provided below to gain better background knowledge on the Pangeo package suite and concepts.

Help Links

Contact

Please feel free to reach out to Sydney Foks (sfoks at usgs.gov) with any questions.