General Information

Basic features of RectiPy

  • Frontend:

    • implement models via the PyRates frontend

    • choose between YAML templates or Python classes to define your RNN

    • choose between various pre-implemented neuron models or implement your custom neuron model

    • add synaptic dynamics and/or delayed coupling

    • implement spiking or rate-based neuron models

    • full control over the RNN parameters that can be trained: Choose between synaptic weights, membrane time constants of neurons, …

    • run pre-implemented training and testing workflows via a single function call, OR use any rectipy.Network as a single unit/layer within your own, custom torch code

    • track any state variable of your model with any temporal resolution during training/testing/simulation procedures

  • Backend:

    • make full use of the PyTorch backend

    • full support of autograd for your parameter optimization

    • use any loss function and optimization algorithm available in torch

    • deploy your model on different hardware

Reference

If you use RectiPY, please cite the most recent release:

https://zenodo.org/badge/523464500.svg

Contact

If you have questions, problems or suggestions regarding RectiPy, please contact Richard Gast.

Contribute

RectiPy is an open-source project that everyone is welcome to contribute to. Check out our GitHub repository for all the source code, open issues etc. and send us a pull request, if you would like to contribute something to our software.