Skip to content

Latest commit

 

History

History
64 lines (43 loc) · 3.51 KB

README.md

File metadata and controls

64 lines (43 loc) · 3.51 KB
Raw
torchHydroNodes Logo torchHydroNodes: A Python package for training and testing hybrid hydrological models combining conceptual/physics-based models and neural networks.

WORK IN PROGRESS / FRAMEWORK UNDER DEVELOPMENT

torchHydroNodes is an evolving Python library designed for developing hybrid hydrological models that integrate conceptual/physics-based models with neural networks. Inspired by Höge et al., 2022, this framework builds upon the original HydroNODE implementation in Julia using the SciML packages.

The library is designed to be modular and user-friendly, enabling users to implement, adapt, and test their own hybrid models with ease. Features include:

  • Support for loading and using datasets in a customizable way.
  • Methods for parameter estimation.
  • Integration of various conceptual models.
  • Compatibility with multiple neural network architectures (e.g., MLP, LSTM, etc.).
  • Flexible hybrid coupling strategies.
  • Options for training single- and multi-basin models.
  • A variety of metrics for model training and evaluation.
  • Access to multiple ODE solvers leveraging SciPy and torchdiffeq.

The source code structure, data handling, and metrics implementation are inspired by the NeuralHydrology library.

Comprehensive documentation, including a getting started guide, configuration parameter details, simulation setup instructions, model descriptions, use cases, and tutorials, is available at: torchhydronodes.readthedocs.io.

Key Features

  • Modular Design: A flexible framework to implement, adapt, and extend hybrid hydrological models for various use cases.
  • Neural Network Integration: Native support for MLP, LSTM, and other neural architectures.
  • Dataset Handling: Tools to load and preprocess multiple hydrological datasets with minimal setup.
  • Hybrid Coupling: Seamless integration of conceptual models with neural networks for improved accuracy and interpretability.
  • ODE Solvers: Access to SciPy's and Torchdiffeq’s ODE solvers for efficient numerical integration.
  • Metrics: Comprehensive metrics tailored for hydrological simulations, enabling detailed performance evaluations.

License

torchHydroNodes is developed by the Numerical Simulation Research Lab at the Department of Computer Science, University of Saskatchewan. It is distributed under the BSD 3-Clause License. Please see the LICENSE file for specific details.

Feel free to contribute, explore the documentation, or join the discussion to help improve and expand torchHydroNodes!