TorchDyno is a PyTorch-based library for the implementation of dynamical systems. It provides a simple and flexible way to build and train non-standard recurrent models, such as Reservoir Computing models. TorchDyno is designed to be easy to use, efficient, and flexible. It is built on top of PyTorch, which makes it easy to integrate with other PyTorch-based libraries.

Here's the link to the documentation: https://torchdyno.readthedocs.io/en/latest/

Installation

TorchDyno depends on PyTorch but does not pin a specific build — install the torch (and, for the bundled vision dataset, torchvision) wheel that matches your platform/accelerator first, following the official instructions.

Then install TorchDyno:

pip install torchdyno

The bundled example datasets that need extra dependencies (SequentialMNIST → torchvision, HHAR → pandas) are available via the datasets extra:

pip install "torchdyno[datasets]"

Development

TorchDyno uses uv:

uv sync --all-extras --group dev
uv run pytest

Credits

We thank eclypse-org and Jacopo Massa for the structure and the template of the documentation!