Open-Awesome
CategoriesAlternativesStacksSelf-HostedExplore
Open-Awesome

© 2026 Open-Awesome. Curated for the developer elite.

TermsPrivacyAboutGitHubRSS
  1. Home
  2. Common Lisp
  3. MGL

MGL

MITCommon Lisp

A Common Lisp machine learning library focusing on neural networks, Boltzmann machines, and Gaussian processes with BLAS and CUDA support.

GitHubGitHub
651 stars39 forks0 contributors

What is MGL?

MGL is a comprehensive machine learning library for Common Lisp, primarily focusing on neural networks including backpropagation networks, Boltzmann machines, and Gaussian processes. It emphasizes performance and power over ease of use, leveraging BLAS and CUDA for efficient computation. The library provides low-level control and flexibility for advanced machine learning tasks, with built-in support for dataset utilities, gradient-based optimization, monitoring, and model persistence.

Target Audience

Common Lisp developers and researchers who require high-performance, low-level control over neural network architectures and other machine learning models, particularly those working on tasks like regression, classification, and unsupervised learning. It is suited for users who prioritize computational efficiency and are comfortable with a more complex, flexible API.

Value Proposition

Developers choose MGL for its focus on performance via BLAS and CUDA integration, its comprehensive support for various neural network types (feed-forward, recurrent, Boltzmann machines), and its emphasis on low-level control and flexibility. Unlike higher-level libraries, MGL offers advanced features like batch processing, gradient-based optimization algorithms, and extensive monitoring tools, making it ideal for custom or research-oriented machine learning projects in Common Lisp.

Overview

Common Lisp machine learning library.

Use Cases

Best For

  • Implementing and training backpropagation neural networks (feed-forward and recurrent) in Common Lisp with support for various activation functions and loss functions.
  • Building and experimenting with Boltzmann machines for unsupervised learning tasks in Common Lisp.
  • Applying Gaussian processes for non-parametric Bayesian modeling in regression and classification problems within Common Lisp.
  • Performing gradient-based optimization (e.g., gradient descent, conjugate gradient) with custom optimizers in Common Lisp machine learning workflows.
  • Conducting resampling tasks like cross-validation, bagging, and stratified sampling for model evaluation and selection in Common Lisp.
  • Monitoring and evaluating machine learning models with built-in tools for tracking training statistics, classification metrics, and confusion matrices in Common Lisp.

Not Ideal For

  • Projects requiring quick prototyping with minimal setup and configuration due to manual installation of dependencies like CL-CUDA and BLAS libraries.
  • Teams without expertise in Common Lisp or low-level machine learning concepts, as the library emphasizes power over ease of use.
  • Applications needing out-of-the-box support for modern deep learning architectures like convolutional neural networks or transformers, which are not mentioned in the documentation.
  • Environments where Python or other mainstream ML ecosystems are preferred for community support, tooling, and integration with popular frameworks.

Pros & Cons

Pros

High-Performance Computation

Leverages BLAS and CUDA for efficient matrix operations, with automatic fallback to CPU if GPU is unavailable, ensuring performance across hardware setups.

Comprehensive Neural Network Support

Implements backpropagation networks (feed-forward and recurrent), Boltzmann machines, and Gaussian processes, covering a wide range of machine learning tasks from supervised to unsupervised learning.

Advanced Monitoring and Evaluation

Provides built-in monitors, measurers, and counters for tracking training statistics, classification metrics, and confusion matrices, facilitating detailed model debugging and analysis.

Flexible Dataset Handling

Includes utilities for data partitioning, cross-validation, bagging, and stratified sampling, enabling robust model evaluation and resampling techniques.

Cons

Complex Setup and Dependencies

Requires manual installation of dependencies like CL-CUDA and MGL-MAT, which are not in Quicklisp, and configuration of BLAS/CUDA libraries, making initial setup cumbersome.

Niche Language and Ecosystem

Built for Common Lisp, which has a smaller community and fewer integrations with mainstream ML tools, limiting access to resources, tutorials, and collaborative support.

Low-Level API with Steep Learning Curve

Emphasizes low-level control and flexibility, requiring users to have deep understanding of machine learning concepts and Common Lisp, which can be barrier to entry for rapid development.

Frequently Asked Questions

Quick Stats

Stars651
Forks39
Contributors0
Open Issues5
Last commit4 months ago
CreatedSince 2014

Tags

#blas#cuda#gradient-descent#neural-networks#cross-validation#machine-learning#common-lisp#backpropagation#gaussian-processes

Built With

B
BLAS
C
CL-CUDA
C
CUDA
M
MGL-PAX
C
Common Lisp

Included in

Common Lisp2.9k
Auto-fetched 13 hours ago

Related Projects

clmlclml

Common Lisp Machine Learning Library

Stars268
Forks35
Last commit4 years ago
llama.clllama.cl

Inference Llama in Common Lisp

Stars66
Forks3
Last commit1 year ago
authorauthor

to win the Higgs Boson Machine Learning Challenge

Stars0
Forks0
Last commit
AI ChallengeAI Challenge

in 2010 using Common Lisp, but without MGL, as no machine learning was needed. A related talk (59', 2013)

Stars0
Forks0
Last commit
Community-curated · Updated weekly · 100% open source

Found a gem we're missing?

Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.

Submit a projectStar on GitHub