A project to implement all 80+ scipy continuous probability distributions in PyTorch.
This project was undertaken to properly implement a Hamiltonian Monte Carlo in the PyLevyProcess library using hamiltorch. The existing PyLevyProcess implementation uses a direct Monte Carlo sampler. For more rigorous returns innovation sampling, Hamiltonian Monte Carlo is needed to ensure no/little autocorrelation among sampled returns, as needed for a Levy Process. As neither scipy nor Jax support gradients for all probability distributions, this project fills a vital niche.
You can install the package using pip:
pip install git+https://github.com/datstat-consulting/PyTorchDistributionsExtended
WIP
WIP