📊 Weights & Biases (W&B) Flashcards
Experiment Tracking and Model Management for ML Teams
📌 What is W&B?
Weights & Biases is a tool for tracking experiments, visualizing metrics, managing models, and collaborating on ML workflows.
⚙️ Setup
Install via pip install wandb and use wandb.init() to begin logging an experiment.
📈 Metrics Logging
Use wandb.log() to track training/validation metrics in real time across epochs and batches.
🔍 Hyperparameter Tracking
Automatically track and compare hyperparameters used in different runs for optimization and reproducibility.
📂 Project & Run Organization
Organize experiments by project and run with a web dashboard that supports filtering, comparisons, and notes.
🎯 W&B Sweeps
Automate hyperparameter optimization using W&B Sweeps with Bayesian search, grid, or random strategies.
🧠 Model Management
Use W&B Artifacts to version and manage datasets, models, and evaluation results collaboratively.
🧪 Dataset Versioning
Track and version datasets like Git for data using Artifacts, with full lineage and metadata.
📺 Visualization Dashboards
Interactive plots, custom charts, and parallel coordinates for hyperparameter sweep analysis.
🌐 Team Collaboration
Share live reports, dashboards, and analysis with team members and stakeholders across projects.
⚡ Auto-Logging
Out-of-the-box support for popular ML libraries like Keras, PyTorch Lightning, Sklearn, Hugging Face, XGBoost, and more.
🛠️ Integration Ecosystem
Seamlessly integrates with Colab, Jupyter, GitHub, S3, GCP, AWS, and other parts of the ML stack.