Skip to content

Project Structure

Codebase Organization

Understanding the TorchEBM project structure helps you navigate the codebase and contribute effectively. This guide provides an overview of the repository organization.

Repository Overview

The TorchEBM repository is organized as follows:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
torchebm/
├── torchebm/              # Main package source code
│   ├── core/              # Core functionality and base classes
│   ├── samplers/          # Sampling algorithms implementation
│   ├── losses/            # BaseLoss functions for training
│   ├── models/            # Neural network model implementations
│   ├── cuda/              # CUDA optimized implementations
│   └── utils/             # Utility functions and helpers
├── tests/                 # Test directory
├── docs/                  # Documentation
├── examples/              # Example applications
├── benchmarks/            # Performance benchmarks
├── setup.py               # Package setup script
└── README.md              # Project README

Main Package Structure

torchebm.core

Contains the core functionality including base classes and essential components:

  • base.py - Base classes for the entire library
  • energy_function.py - Base energy function class and interface
  • analytical_functions.py - Analytical energy function implementations
  • distributions.py - Probability distribution implementations

torchebm.samplers

Implementations of various sampling algorithms:

  • base.py - Base sampler class
  • langevin_dynamics.py - Langevin dynamics implementation
  • hmc.py - Hamiltonian Monte Carlo
  • metropolis_hastings.py - Metropolis-Hastings
  • [Other sampler implementations]

torchebm.losses

BaseLoss functions for training energy-based models:

  • base.py - Base loss class
  • contrastive_divergence.py - Contrastive divergence implementations
  • score_matching.py - Score matching methods
  • [Other loss implementations]

torchebm.models

Neural network model implementations:

  • mlp.py - Multi-layer perceptron energy models
  • cnn.py - Convolutional neural network energy models
  • ebm.py - Generic energy-based model implementations
  • [Other model architectures]

torchebm.cuda

CUDA-optimized implementations for performance-critical operations:

  • kernels/ - CUDA kernel implementations
  • bindings.cpp - PyTorch C++ bindings
  • ops.py - Python interfaces to CUDA operations

torchebm.utils

Utility functions and helpers:

  • device.py - Device management utilities
  • visualization.py - Visualization tools
  • data.py - Data loading and processing utilities
  • logging.py - Logging utilities

Tests Structure

The tests directory mirrors the package structure:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
tests/
├── unit/                  # Unit tests
│   ├── core/              # Tests for core module
│   ├── samplers/          # Tests for samplers module
│   ├── losses/            # Tests for losses module
│   └── utils/             # Tests for utilities
├── integration/           # Integration tests
├── performance/           # Performance benchmarks
├── conftest.py            # Pytest configuration and fixtures
└── utils.py               # Test utilities

Documentation Structure

The documentation is built with MkDocs and organized as follows:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
docs/
├── index.md               # Home page
├── getting_started.md     # Quick start guide
├── guides/                # User guides
├── api/                   # API reference (auto-generated)
├── examples/              # Example documentation
├── developer_guide/       # Developer documentation
│   ├── contributing.md    # Contributing guidelines
│   ├── code_style.md      # Code style guide
│   └── ...                # Other developer docs
└── assets/                # Static assets
    ├── images/            # Images
    ├── stylesheets/       # Custom CSS
    └── javascripts/       # Custom JavaScript

Examples Structure

Example applications to demonstrate TorchEBM usage:

1
2
3
4
examples/
├── basic/                 # Basic usage examples
├── advanced/              # Advanced examples
└── real_world/            # Real-world applications

Dependencies and Requirements

TorchEBM has the following dependencies:

  • PyTorch

    Primary framework for tensor operations and automatic differentiation.

  • NumPy

    Used for numerical operations when PyTorch isn't needed.

  • Matplotlib

    For visualization capabilities.

  • tqdm

    For progress bars during long operations.

Entry Points

The main entry points to the library are:

  • torchebm.core - Import core functionality
  • torchebm.samplers - Import samplers
  • torchebm.losses - Import loss functions
  • torchebm.models - Import neural network models

Example of typical import patterns:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
# Import energy_functions components
from torchebm.core import BaseEnergyFunction, GaussianEnergy

# Import samplers
from torchebm.samplers import LangevinDynamics, HamiltonianMC

# Import loss functions
from torchebm.losses import ContrastiveDivergence

# Import utilities
from torchebm.utils import visualize_samples

Package Management

TorchEBM uses setuptools for package management with setup.py:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
# setup.py excerpt
setup(
    name="torchebm",
    version=__version__,
    description="Energy-Based Modeling library for PyTorch",
    packages=find_packages(),
    install_requires=[
        "torch>=1.9.0",
        "numpy>=1.20.0",
        "matplotlib>=3.4.0",
        "tqdm>=4.60.0",
    ],
    # Additional configuration...
)

Version Control Structure

  • We follow Conventional Commits for our commit messages
  • Feature branches should be named feature/feature-name
  • Bugfix branches should be named bugfix/bug-name
  • Release branches should be named release/vX.Y.Z

Finding Your Way Around

When contributing to TorchEBM, start by exploring the relevant directory for your feature. For example, if you're adding a new sampler, look at the existing implementations in torchebm/samplers/ to understand the pattern.