Hyperparameter Tuning

Finding the optimal set of hyperparameters is one of the most critical steps in achieving peak performance with advanced forecasting models. The fusionlab-learn library provides a powerful and flexible tuning framework built on top of the industry-standard Keras Tuner library.

Our tuning utilities are designed to automate the search for the best model architecture and training configurations, saving you significant time and effort. This section provides detailed guides and practical examples for each of the available tuners.

The guides are organized by the model families they are designed to optimize.

PINN Model Tuning

• Hyperparameter Tuning with HydroTuner
  • Core Concepts
  • End-to-End Workflow
  • API Reference
• HydroTuner: Usage Examples
  • Example 1: A Comprehensive Workflow for Tuning TransFlowSubsNet
  • Example 2: Tuning PIHALNet

Hybrid & Transformer Model Tuning

• Tuning HALNet with the XTFTTuner
  • End-to-End Workflow
• TFT Forecast Tuner Guide
  • Prerequisites
  • Choosing Your Tuning Approach: Functions vs. Classes
  • Tuner Functions (Legacy Approach)
  • Internal Model Builder
• TFT & XTFT Tuning Examples
  • Prerequisites
  • Common Setup for All Examples
  • Example 1: Tuning XTFT for Quantile Forecasting
  • Tuning Standard TFT Variants
• XTFT Tuning
  • Prerequisites

Other Guides & Examples

• Class-Based Tuner Guide
  • Prerequisites
  • The Class-Based Approach
  • Customizing the Tuning
• Example: Tuning PIHALNet with PIHALTuner
  • Prerequisites
  • Section 1: Tuning with Synthetic Data
  • Section 2: Real Application Case - Example Workflow

Legacy Tuners

• Hyperparameter Tuning (Legacy PiHALTuner)
  • Core Concepts
  • End-to-End Workflow
  • API Reference

Note

The tuning examples use small search spaces and few trials for demonstration purposes. For real-world applications, you’ll likely want to explore a wider range of hyperparameters and run the tuner for more trials and epochs to find the best configurations for your specific dataset and task.