“This session will introduce state-of-the-art time series forecasting techniques using PyTorch and PyMC3. Sales and demand forecasting is a highly value-adding application of forecasting. This use case will be demonstrated using non-hierarchical methods such as LSTMs, Temporal Fusion Transformers, NeuralProphet and DeepAR with PyTorch, as well as hierarchical forecasting methods such as mixed-effect models using PyMC3. Participants will also learn about feature engineering methods and ML deployment methods for forecasting popularly used in the industry.
Key Takeaways:
- Feature Engineering for Forecasting using tsfresh library
- Demand Forecasting with PyTorch
- Long Short-term Memory Networks (LSTMs)
- Temporal Fusion Transformers
- NeuralProphet by Meta AI
- DeepAR
- Hierarchical Demand Forecasting with PyMC3
- Mixed-effect models
- ML Deployment for Forecasting
- SHAP for interpretability
- Monitoring ex-ante vs. ex-post forecasts”
