Weights & Biases

Weights & Biases

In Deepnote, Weights & Biases (W&B) serves as a tool for tracking, visualizing, and improving machine learning models. Its easy integration into the workflow allows developers to monitor experiment results, track hyperparameters, and visualize training metrics in real time. It's a platform that speeds up machine learning development by helping developers:

• Track experiments in real time to monitor performance over iterations.

• Version datasets for consistent and reliable experiments.

• Evaluate performance with detailed visualizations of metrics like accuracy and loss.

• Reproduce models by logging configurations, environments, and code.

• Spot regressions to quickly identify and resolve performance dips.

• Share insights easily with teammates or the community.

W&B covers a wide range of resources in its Tutorials section for those interested in learning more about how to improve their machine learning workflows.

Using W&B in Deepnote

Wandb (W&B)can be easily integrated into Deepnote by following steps:

!pip install --upgrade wandb
!wandb login <your_wandb_token>

W&B token can be generated by visiting the Weights & Biases login page once logged in. Token can be copied from your W&B account settings and used in Deepnote's terminal. The same steps can be done directly through Deepnote's terminal, making it easy to set up and get started.

For illustration, an example of using W&B in Deepnote can be this project: Predict Olympic medals in PyTorch, where a ML model was trained using PyTorch or Predict Olympic medals in SciKit learn.

If the notebook/data app is public, the Wandb's token shouldn't be shared publicly. In Deepnote, it can be soloved with the Integrations section in the left sidebar, when select the option Create new integration, then by clicking on Environment variables to create a variable and insert the token generated in the W&B account.

In the image below, the example workspace of a specific project is shown. The graph displays the progression of metrics such as test accuracy or loss. The outputs and logs are also available in the wandb folder directly within the Deepnote project in the left panel.

Conclusion

Weights & Biases (W&B) is a powerful tool that makes tracking experiments, evaluating models, and collaborating easier for machine learning projects. Its seamless integration with Deepnote gives developers everything they need to build and improve models efficiently. From easy setup to advanced visualization tools, W&B helps you create better models, faster.