A Library for Modelling Probabilistic Hierarchical Graphical Models in PyTorch

Last update: Nov 28, 2022

Overview

README of "PyTorch-ProbGraph"

What is PyTorch-ProbGraph?

PyTorch-ProbGraph is a library based on amazing PyTorch (https://pytorch.org) to easily use and adapt directed and undirected Hierarchical Probabilistic Graphical Models. These include Restricted Boltzmann Machines, Deep Belief Networks, Deep Boltzmann Machines and Helmholtz Machines (Sigmoid Belief Networks).

Models can be set up in a modular fashion, using UnitLayers, layers of Random Units and Interactions between these UnitLayers. Currently, only Gaussian, Categorical and Bernoulli units are available, but an extension can be made to allow all kinds of distributions from the Exponential family. (see https://en.wikipedia.org/wiki/Exponential_family)

The Interactions are usually only linear for undirected models, but can be built from arbitrary PyTorch torch.nn.Modules (using forward and the backward gradient).

There is a pre-implemented fully-connected InteractionLinear, one for using existing torch.nn.Modules and some custom Interactions / Mappings to enable Probabilistic Max-Pooling. Interactions can also be connected without intermediate Random UnitLayers with InteractionSequential.

This library was built by Korbinian Poeppel and Hendrik Elvers during a Practical Course "Beyond Deep Learning - Uncertainty Aware Models" at TU Munich. Disclaimer: It is built as an extension to PyTorch and not directly affiliated.

Documentation

A more detailed documentation is included, using the Sphinx framework. Go inside directory 'docs' and run 'make html' (having Sphinx installed). The documentation can then be found inside the _build sub-directory.

Examples

There are some example models, as well as an evaluation script using the EMNIST dataset in the examples folder.

License

This library is distributed in a BSD 3-clause license.

Setup

The library is accessible via the PyPi repository and can be install by: pip install pytorch_probgraph

References

Ian Goodfellow and Yoshua Bengio and Aaron Courville, http://www.deeplearningbook.org

Jörg Bornschein, Yoshua Bengio Reweighted Wake-Sleep https://arxiv.org/abs/1406.2751

Geoffrey Hinton, A Practical Guide to Training Restricted Boltzmann Machines https://www.cs.toronto.edu/~hinton/absps/guideTR.pdf

Ruslan Salakhutdinov, Learning Deep Generative Models https://tspace.library.utoronto.ca/handle/1807/19226

Honglak Lee et al., Convolutional Deep Belief Networks for Scalable Unsupervised Learning of Hierarchical Representations, ICML09

G.Hinton, S. Osindero A fast learning algorithm for deep belief nets

You might also like...

Scikit-learn compatible estimation of general graphical models

A Library for Modelling Probabilistic Hierarchical Graphical Models in PyTorch

Related tags

Overview

README of "PyTorch-ProbGraph"

What is PyTorch-ProbGraph?

Documentation

Examples

License

Setup

References

You might also like...

Scikit-learn compatible estimation of general graphical models

Simple PyTorch hierarchical models.

Adversarial Attacks on Probabilistic Autoregressive Forecasting Models.

Denoising Diffusion Probabilistic Models

ILVR: Conditioning Method for Denoising Diffusion Probabilistic Models (ICCV 2021 Oral)

Topic Modelling for Humans

Civsim is a basic civilisation simulation and modelling system built in Python 3.8.

Dataloader tools for language modelling

Trans-Encoder: Unsupervised sentence-pair modelling through self- and mutual-distillations

Releases(v0.1-beta)

v0.1-beta(Aug 7, 2020)

Owner

Korbinian Pöppel

Code for Phase diagram of Stochastic Gradient Descent in high-dimensional two-layer neural networks

Video Representation Learning by Recognizing Temporal Transformations. In ECCV, 2020.

Chess reinforcement learning by AlphaGo Zero methods.

Official code of the paper "ReDet: A Rotation-equivariant Detector for Aerial Object Detection" (CVPR 2021)

Implementation of popular SOTA self-supervised learning algorithms as Fastai Callbacks.

Original code for "Zero-Shot Domain Adaptation with a Physics Prior"

GANmouflage: 3D Object Nondetection with Texture Fields

Converts given image (png, jpg, etc) to amogus gif.

KwaiRec: A Fully-observed Dataset for Recommender Systems (Density: Almost 100%)

Parasite: a tool allowing you to compress and decompress files, to reduce their size

Banglore House Prediction Using Flask Server (Python)

Find the Heart simple Python Game

PySOT - SenseTime Research platform for single object tracking, implementing algorithms like SiamRPN and SiamMask.

PoseCamera is python based SDK for human pose estimation through RGB webcam.

Harmonious Textual Layout Generation over Natural Images via Deep Aesthetics Learning

Home repository for the Regularized Greedy Forest (RGF) library. It includes original implementation from the paper and multithreaded one written in C++, along with various language-specific wrappers.

MNE: Magnetoencephalography (MEG) and Electroencephalography (EEG) in Python

Non-Official Pytorch implementation of "Face Identity Disentanglement via Latent Space Mapping" https://arxiv.org/abs/2005.07728 Using StyleGAN2 instead of StyleGAN

Offcial implementation of "A Hybrid Video Anomaly Detection Framework via Memory-Augmented Flow Reconstruction and Flow-Guided Frame Prediction, ICCV-2021".

Pathdreamer: A World Model for Indoor Navigation