Lab 4: Experiment Management
In this lab, we'll work through an entire experiment management workflow for model development, using a tool called Weights & Biases.
Outline
- 00:00 Why do we need experiment management?
- 02:24 Tracking experiments with TensorBoard
- 04:16 Experiment management with Weights & Biases
- 06:48 A guided tour of the W&B run interface
- 12:12 Exploratory data analysis with W&B Tables
- 14:00 Project management with W&B
- 16:27 Artifact versioning with W&B
- 18:52 Programmatic API access to W&B
- 20:14 Collaboration tools in W&B
- 25:00 Hyperparameter sweeps in W&B
- 28:15 Overview of exercises
Wait, what happened to labs 1 through 3?
The first three labs review some pre-requisites for the course -- DNN architectures and the basics of model training.
You can find them here.
If you're already basically familiar with training neural networks in any framework, you really only need to review Lab 02a, on using PyTorch Lightning.