Pinterest Tutorial: Logging Your First Run to Weights & Biases
Created on October 3|Last edited on October 3
Comment
Introduction
This tutorial will teach you how to track your experiments on Weights & Biases using MLEnv and get familiar with the Weights & Biases UI. It is broken down into the following sections:
Weights & Biases Configurations: we will look at some useful W&B configurations that can help streamline your model development process.
Log a Run to Weights & Biases: log your first run to W&B using MLEnv and cover the basics of navigating the UI.
Hyperparameter Tuning with Weights & Biases: we will use hyperparameter tuning to learn about grouped runs, comparing runs and customizing your workspace.
Creating a Report: embed visualizations, document your learning, and share with collaborators in a report like this.
Prep Work [5 mins]
New Weights & Biases users need to sign up and store their API key in Knox for the first time. It takes some time for Knox to update so it is recommended you follow these steps before the tutorial.
Sign up on Weights & Biases
New users need to first create an account on Weights & Biases. In Okta, click on the Weights & Biases tile. This should lead you to the sign up page with your information prepopulated. Verify the information looks correct and sign up.
Store your API Key
Note: If you didn’t do this step before the tutorial, we can skip this for now and pass the api key directly to the launcher script with
--trainer_config.logger_config.wandb_config.api_key=<your-api-key>.
We use Knox to store your Weights & Biases API key. Run this script on your devapp to store your API key in Knox and grant appropriate access.
> cd ~/code/pinboard/machine-learning> ./scripts/bin/idoc "../pinboard/ml_platform/scripts/wandb/store_entity_key.py -u <ldap> -k <api_key>"
Tutorial
Add a comment