modflow

This is our modding of the nflows package to implement a stricter identity init and avoid numerical errors.

Install

To use it clone the repo, cd ./modflow and the

pip install -e modflow

or if using from a colab you should be able to pip install the main branch as

pip install "modflow @ git+https://github.com/cms-flashsim/modflow"

Use

At the moment we have modded three classes of transform. They are the AutoregressiveAffine, AutoregressiveRQSpline, CouplingRQSpline. To use them install the package and then do:

from modflow.transformers import (MaskedAffineAutoregressiveTransform,
MaskedPiecewiseRationalQuadraticAutoregressiveTransform,
PiecewiseRationalQuadraticCouplingTransform)

Compared to nflows they have only one additional input argument init_identity which is True by default. Note that the names are the same as nflows so if you need both maybe use a import xyz as jkl syntax.