https://github.com/joeybose/HyperbolicNF

ICML 2020 Paper: Latent Variable Modelling with Hyperbolic Normalizing Flows
https://github.com/joeybose/HyperbolicNF

Last synced: 3 months ago
JSON representation

ICML 2020 Paper: Latent Variable Modelling with Hyperbolic Normalizing Flows

• Host: GitHub
• URL: https://github.com/joeybose/HyperbolicNF
• Owner: joeybose
• Created: 2020-06-08T20:02:55.000Z (over 4 years ago)
• Default Branch: master
• Last Pushed: 2022-12-08T04:20:39.000Z (almost 2 years ago)
• Last Synced: 2024-07-04T01:02:16.116Z (4 months ago)
• Language: Python
• Homepage:
• Size: 20.2 MB
• Stars: 54
• Watchers: 3
• Forks: 7
• Open Issues: 23
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Normalizing Flows for Hyperbolic Spaces and Beyond!

![alt text](https://github.com/joeybose/HyperbolicNF/blob/master/hyperflow_animation_large.gif "Hyperbolic NF")

This repository contains code for reproducing results for ICML 2020 paper.

"Latent Variable Modeling with Hyperbolic Normalizing Flows", by:

Avishek Joey Bose, Ariella Smofsky, Renjie Liao, Prakash Panangaden, William L. Hamilton

ArXiv Link: https://arxiv.org/pdf/2002.06336.pdf

If this repository is helpful in your research, please consider citing us.

```

@article{bose2020latent,

  title={Latent Variable Modelling with Hyperbolic Normalizing Flows},

  author={Bose, Avishek Joey and Smofsky, Ariella and Liao, Renjie and Panangaden, Prakash and Hamilton, William L},

  journal={Proceedings of the 37th International Conference on Machine Learning},

  year={2020}

}

```

## Installation

Main Python Packages:

- Pytorch Geometric: https://github.com/rusty1s/pytorch_geometric

Follow the installation instructions carefully for this package! Make sure all

your environment Path variables are exactly as outlined otherwise you will get

weird symbol errors

- Pytorch 1.5

- WandB for logging

Other packages can be found in Requirements.txt but not all from that list are needed.

Download the datasets:

`python -m data.download`

## Running Hyperbolic VAE

`python main.py --dataset=mnist --batch_size=100 --epochs=100 --model=hyperbolic --wandb --namestr="MNIST 2-HyperbolicVAE"`

## Running Euclidean Flow

`python main.py --dataset=mnist --batch_size=100 --epochs=100 --model=euclidean --flow_model=RealNVP --wandb --namestr="MNIST 2-Hyperbolic 2-RealNVP"`

## Running Flow Hyperbolic VAE

`python main.py --dataset=mnist --batch_size=100 --epochs=100 --model=hyperbolic --flow_model=TangentRealNVP --n_blocks=4 --wandb --namestr="MNIST 2-Hyperbolic 4-TangentRealNVP"`

## Reference code repos

1. "A Wrapped Normal Distribution on Hyperbolic Space for Gradient-Based

   Learning": https://github.com/pfnet-research/hyperbolic_wrapped_distribution

2. "Mixed-Curvature Variational Autoencoder":

   https://www.dropbox.com/s/tzilf229js1gsqu/mvae.zip?dl=0

3. "Hyperbolic Neural Networks": https://github.com/dalab/hyperbolic_nn

4. "Hyperbolic Graph Convolutional Neural Networks": https://github.com/HazyResearch/hgcn