Extensions to torch distributions

Examples:

import torch.distributions as torchd
import probabll.distributions as probd
probd.Kumaraswamy(a=torch.ones(1), b=torch.ones(1))
probd.MixtureOfGaussians(logits=torch.ones(2), components=torchd.Normal(loc=torch.zeros(2), scale=torch.ones(2)))

Installation

• Dependencies

pip install -r requirements.txt

• For developers

python setup.py develop

• For users

pip install .

Sparse relaxations to binary random variables

Sparse distributions for pytorch:

• HardKumaraswamy
• Mixture of delta and Exponential

Contributors

Though the history of commits won't necessarily show it, this code base received contributions from:

• Nicola De Cao
• Jasmijn Bastings
• Bryan Eikema
• Wilker Aziz

Citation

If you use HardKumaraswamy please cite this paper:

• Jasmijn Bastings, Wilker Aziz and Ivan Titov. In ACL, 2019.