Hands-On Exercises

Welcome to the exercises section of the fusionlab-learn user guide. This collection of hands-on tutorials provides practical, step-by-step walkthroughs for the library’s key features, data preparation workflows, and model architectures.

Whether you are getting started with basic forecasting or implementing an advanced, physics-informed model, these guides are designed to help you master the library’s capabilities. We recommend starting with the Foundational Skills if you are new to the library.

Foundational Skills

• Exercise: Basic Point Forecasting with Flexible TFT
  • Prerequisites
  • Discussion of Exercise
• Exercise: Data Preparation Workflow for Case History Data
  • Prerequisites
  • Discussion of Exercise
• Exercise: Quantile Forecasting with TFT Variants
  • Prerequisites
  • Exercise 1: Quantile Forecasting with Flexible TemporalFusionTransformer
  • Exercise 2: Quantile Forecasting with Stricter TFT
  • Discussion of Exercise

Data-Driven Forecasting Models

• Exercise: Advanced Forecasting with BaseAttentive
  • Prerequisites
  • Step 1: Imports and Setup
  • Step 2: Generate Synthetic Data
  • Step 3: Preprocess and Reshape Data
  • Step 4: Define, Compile, and Train the Model
  • Step 5: Visualize the Probabilistic Forecast
  • Discussion of Exercise
• Exercise: Forecasting with HALNet (All Inputs Required)
  • Prerequisites
  • Discussion of Exercise
• Exercise: Forecasting with a Pure Transformer
  • Prerequisites
  • TimeSeriesTransformer
• Exercise: Forecasting with Stricter TFT (All Inputs Required)
  • Prerequisites
  • Discussion of Exercise
• Exercise: Advanced Quantile Forecasting with XTFT
  • Prerequisites
• Exercise: Advanced Forecasting with SuperXTFT
  • Prerequisites
  • Discussion of Exercise

Physics-Informed Models (PINNs)

• Exercise: Hybrid Forecasting with PIHALNet
  • Prerequisites
  • Discussion of Exercise
• Exercise: Hybrid Forecasting with TransFlowSubsNet
  • Prerequisites
  • Discussion of Exercise
• Exercise: Physics-Informed Forecasting with PIHALNet & TransFlowSubsNet
  • Prerequisites
  • Step 0: Preamble & Configuration
  • Step 1: Load and Inspect the Dataset
  • Step 2: Preprocessing - Feature Selection & Cleaning
  • Step 3: Preprocessing - Encoding & Scaling
  • Step 4: Define Feature Sets & Generate Sequences
  • Step 5: Create tf.data.Dataset
  • Step 6: Model Training
  • Step 7: Visualize Results
  • Discussion of Exercise
• Exercise: Solving a Forward Problem with PiTGWFlow
  • Prerequisites
  • Discussion of Exercise
• Exercise: A Basic PIHALNet Forecasting Workflow
  • Prerequisites
  • Discussion of Exercise

Advanced Topics & Tools

• Exercise: Hyperparameter Tuning with HydroTuner
  • Prerequisites
• Exercise: Anomaly Detection
  • Prerequisites
  • Exercise 1: Unsupervised Anomaly Detection with LSTM Autoencoder
  • Exercise 2: Using SequenceAnomalyScoreLayer (Conceptual)
  • Exercise 3: Using PredictionErrorAnomalyScore