FR: Volume Preserving Modification

[robsalomone]

Hi Edvard,

I was wondering if it is possible for you to add an option to make your MAF implementation be a volume preserving transform? I need this for a specific purpose, and it should be easy but I don't know your code well enough.

All one needs to do is make sure the last scale variable is the negative sum of the others. That way the log-determinant is always zero.

The idea is done with Real NVP in the paper here, but it is very simple:
https://arxiv.org/abs/2001.04872

Much appreciated if you can make this modification!