README.md

Weights & Biases 101

Course info

• Course url
• code repository

Course contents

1. Introducing W & B
2. Instrumenting W&B in your code
3. Exploring W&B workspace
4. Comparing & analyzing experiments
5. Using W&B beyond experiment tracking
6. Testing your knowledge
7. Logging your first run
8. More resources for you

Introducing W & B

• This lesson introduces Weights & Biases (W&B)
  • An MLOps platform to track and organize machine learning experiments.
• Explains how W&B can be integrated to log metrics and configurations for easy tracking and visualization.
  • Demonstrated using a PyTorch training script.

Instrumenting W&B in your code

• Demonstrates how to install and integrate W&B Python client into training script.
• Shows how to refactor code to create a run, gather configurations and pass them to wandb.init().
• wandb.log() to store and visualize the history of each metric throughout the training process.

Exploring W&B workspace

• Instructor demonstrates
  • How to run a training script integrated with Weights & Biases.
  • Navigate the W&B workspace to view real-time metrics, interact with plots, and access system metrics.
  • Highlights W&B's ability to capture information that helps with reproducibility.

Comparing & analyzing experiments

• Proper way of finishing W&B run:
  • Python script:
    • Finishes as and when script run is over.
  • Notebook:
    • wandb.finish() or
    • Context manager

Using W&B beyond experiment tracking

• Covers the advanced features of the Weights & Biases platform.
  • Sweeps
    • Hyperparameter optimization
  • Tables
    • Data visualization and evaluation
  • Artifacts
    • Data lineage tracking
  • Models
    • Storing and reproducing model development
  • Reports
    • Sharing key metrics and analysis with stakeholders.

Testing your knowledge

Logging your first run

• Example dashboard
• User settings
• Colab Notebooks
  • Intro to Weights & Biases
  • Hyperparameters sweeps using PyTorch

More resources for you

• W&B Fully Connected
  • Blog on ML and MLOps knowledge
• W&B Docs
• W&B community
• Youtube: Tune Hyperparameters Easily with W&B Sweeps
• W&B Server
  • Basic Setup describes steps to host locally.

Assignments

Assignment	Description
Intro to W&B	Run an experiment by passing hyperparameters. Log metrics. Visualize results in dashboard.
Simple PyTorch Integration	Extension of above assigment by using CNN model. Save model in ONNX format.

Certificate

• Course completion certificate