A PyTorch implementations of Masked Autoregressive Flow and some other invertible transformations from Glow: Generative Flow with Invertible 1x1 Convolutions and Density estimation using Real NVP.

For MAF, I'm getting results similar to ones reported in the paper. GLOW requires some work.

python main.py --dataset POWER

Available datasets are POWER, GAS, HEPMASS, MINIBONE and BSDS300. For the moment, I removed MNIST and CIFAR10 because I have plans to add pixel-based models later.

The datasets are taken from the original MAF repository. Follow the instructions to get them.

Tests check invertibility, you can run them as

pytest flow_test.py