RBMs

Overview

rbms is a GPU-accelerated package designed to train and analyze Restricted Boltzmann Machines (RBMs). It is intended for students and researchers who need an efficient tool for working with RBMs. Features:

• GPU Acceleration through PyTorch.
• Multiple RBM Types: Supports Bernoulli-Bernoulli RBM and Potts-Bernoulli RBM.
• Extensible Design: Provides an abstract class RBM with methods that can be implemented for new types of RBMs, minimizing the need to reimplement training algorithms, analysis methods, and sampling methods.

Installation

PyPI

rbms can be installed from PyPI:

pip install rbms

The last version on PyPI corresponds to the main branch of this repo. To update the installed package run

pip install rbms --upgrade

Github

If you want a more up-to-date version or wish to contribute, you can install the package directly from this repository:

git clone git@github.com:DsysDML/rbms.git
cd rbms && pip install -e .

If you want to update the package run

git pull

from inside the cloned repo.

Dependencies

The dependencies are included in the pyproject.toml file and should be compiled in the wheel distribution. If you have a GPU, make sure to install PyTorch with GPU support.

Usage

Main Classes and Functions:

• EBM: An abstract class that provides the basic structure and methods for Energy-Based Models (EBMs).
• RBM: An abstract class that provides the basic structure and methods for RBMs.
• BBRBM: A concrete implementation of the Bernoulli-Bernoulli RBM.
• PBRBM: A concrete implementation of the Potts-Bernoulli RBM.
• rbms.sampling: Submodule for sampling methods.
• rbms.training: Submodule for training algorithms.

Basic Example

Train a RBM on MNIST-01 with PCD-100 for 10 000 steps, using 200 hidden nodes and 2000 permanent chains:

rbms train  -d ./data/MNIST.h5 --subset_labels 0 1 \
--num_hiddens 200 --gibbs_steps 100 --num_chains 2000 \
--num_updates 10000 --filename ./RBM_MNIST01.h5

Documentation

The documentation is available at https://dsysdml.github.io/rbms/

Contributing

We welcome contributions to the development of TorchRBM. Here's how you can contribute:

• Fork the Repository: Fork the main repository and make your changes.
• Propose a Merge Request: Once your changes are complete, propose a merge request.
• Testing: Ensure your code is tested using the pytest framework. Include your tests in the tests folder.
• Code Style: Follow the pre-commit configuration for code style.
• Documentation: Document your code using the Google docstring style and provide type hints .
• Dependencies: Avoid adding extra dependencies unless absolutely necessary.

Support

If you encounter any issues, please open an issue on the main repository.

License

rbms is released under the MIT License.