Weights & Biases (WandB)

15.1. Weights & Biases (WandB)#

15.1.1. Wandb#

Weights & Biases (W&B) is a platform designed to streamline experiment tracking and visualization, enabling effective collaboration on large teams. Weights and Biases gives researchers great flexibility in deciding what metrics to log.

15.1.1.1. Logging In#

To get started, first sign up for an account at https://wandb.ai/site/. Next, generate an API key at https://wandb.ai/authorize . Finally, install the wandb package and log in:

pip install wandb
wandb login

Logging in using the W&B Command Line Interface adds your api key to your .netrc file at ~/.netrc (Mac and Linux). Alternatively, you can also manually add your api key to this file following this format:

machine api.wandb.ai
  login user
  password YOUR_API_KEY

15.1.1.2. Simple Tracking#

In your python script, import wandb and start a W&B run object. Set the project name and decide which hyperparameters and metrics to track:

import wandb

run = wandb.init(
    project = "project_name"
    config = {
        "metric_1": metric_1,
        "metric_2": metric_2,
    }
)


for epoch in range(epochs):
    ...
    wandb.log({"accuracy": acc, "loss": loss})

Access your project and view your runs at https://wandb.ai/home. Additionally, invite any collaborators you have to your project at this page. Refer to the quickstart guide for more details.

Contents