MLweb

MLweb is an open-source project that aims at bringing machine learning capabilities to web pages and web applications. See the official website for more information.

MLweb includes the following three components:

• ML.js: a javascript library for machine learning
• LALOLib: a javascript library for scientific computing (linear algebra, statistics, optimization)
• LALOLab: an online Matlab-like development environment (try it at http://mlweb.loria.fr/lalolab/)

Documentation

Documentation for LALOLib and ML.js is available here.

LALOLab comes with an online help including the list of all functions and many examples.

Note to users

This repository is mostly intended for developers wishing to modify or extend these tools. Ready-to-use versions of the tools are available online at:

• https://mlweb.loria.fr/lalolib.js for LALOLib
• https://mlweb.loria.fr/ml.js for ML.js
• https://mlweb.loria.fr/lalolab/ for LALOLab

or as modules (see the documentation for details) at:

• https://mlweb.loria.fr/lalolib-module.min.js
• https://mlweb.loria.fr/mljs-module.min.js
• https://mlweb.loria.fr/lalolib-noglpk-module.min.js
• https://mlweb.loria.fr/mljs-noglpk-module.min.js

Functions provided by LALOLib

• Linear algerbra: basic vector and matrix operations, linear system solvers, matrix factorizations (QR, Cholesky), eigendecomposition, singular value decomposition, conjugate gradient sparse linear system solver, complex numbers/matrices, discrete Fourier transform... )
• Statistics: random numbers, sampling from and estimating standard distributions
• Optimization: steepest descent, BFGS, linear programming (thanks to glpk.js), quadratic programming

See this benchmark for a comparison of LALOLib with other linear algebra javascript libraries.

Machine learning capabilities provided by ML.js

Classification

• K-nearest neighbors,
• Linear/quadratic discriminant analysis,
• Naive Bayes classifier,
• Logistic regression,
• Perceptron,
• Multi-layer perceptron,
• Support vector machines,
• Multi-class support vector machines,
• Decision trees

Regression

• Least squares,
• Least absolute devations,
• K-nearest neighbors,
• Ridge regression,
• LASSO,
• LARS,
• Orthogonal least squares,
• Multi-layer perceptron,
• Kernel ridge regression,
• Support vector regression,
• K-LinReg

Clustering

• K-means,
• Spectral clustering

Dimensionality reduction

• Principal component analysis,
• Locally linear embedding,
• Local tangent space alignment

Installation

Download the source files from here or by cloning this repository and run

cd lalolab
make

to build the libraries in the lalolab/ folder:

lalolib.js and lalolibworker.js  --> for LALOLib
ml.js and mlworker.js            --> for ML.js

Then, you can launch LALOLab by opening lalolab/index.html in a browser, for instance with

firefox index.html

Note to Chrome users: you need to use the --allow-file-access-from-files flag on Chrome command line. For Chromium under Linux, you can use the convenient script lalolab/chromelab.