This repository offers an implementation of some of the most common architectures for normalizing flows. If you are here you may be interested by this nice repository https://github.com/bayesiains/nflows which contains many flow components implementation as well!
pip install git+https://github.com/AWehenkel/Normalizing-Flows/
- torch (> 1.5)
- numpy
- umnn (pip install umnn)
Still in progress, you can check the models folder for basic implementation of affine and monotonic transformations and coupling or autorgressive conditioners.
This one shows how to build simple autoregressive feed-forward normalizing flows.
To come...
To come...
To come...