Perfect tool for ML experiments tracking
Use Cases and Deployment Scope
1. We use Weights and Biases for tracking experiments, metrics, log configs, model artifcats
2. Since our primary work is building ML models we have to track the model metrics to identify where the model is going wrong or how we can improve it / how the model has improved with certain changes.
3. Run hyperparameter sweep and visualize it beautifully on the dashboard. The sweep really helps in finding the best hyperparameter and is very easy to integrate into codebase.
4. Write down report with detailed and interactable charts which helps in comparing experiments and sharing it with public.
Pros
- Metrics Logging
- Hyperparmeters Sweeps
- Model Artifcats
Cons
- Dashboard lags when we log a lot of metrics
- Improved support for matplotlib charts and documentation of wandb custom charts is not straghtforward
Most Important Features
- Charts
- Logging
- Experiment table
- Reports
- Artifcats
Return on Investment
- Made it very easy to track experiments
- Track ML and Business Metrics improvements across experiments
- Reproduce runs which is essential in ML modelling
Other Software Used
Amazon SageMaker, Vertex AI, Google Cloud AI
