matLearn Software Package

matLearn: machine learning algorithm implementations in Matlab

The matLearn package contains Matlab implementations of a wide variety of the most commonly-used machine learning algorithms, all using a simple common interface. It in particular focuses on the following tasks:

Regression: Predict a continuous output variable given observed (continuous or discrete) features. It includes methods that are robust, non-parametric, kernelized, and/or ordinal methods.

Classification: Predict a discrete class label given observed (continuous or discrete) features. It includes methods for binary, multiclass, ordinal, and multi-label settings.

Clustering: Group together similar unlabelled data points. It includes parametric and non-parametric methods.

Dimensionality Reduction: Learn a low-dimensional representation of high-dimensional data while trying to maintain the structure in the higher dimensions.

Density Estimation: Construct an estimate of an underlying probability density function using observations in the data matrix.

Documentation

The documentation for matLearn consists of a set of examples that demonstrate the basic usage of matLearn methods on a variety of synthetic and real datasets. These examples are described below. All matLearn algorithms follow a common method interface:

Supervised Interface:

model = matLearn_<TYPE>_<NAME>(X, y, options)

X: n-by-d design matrix
y: n-by-1 target (labels) vector
options: additional fields specified by individual function documentation
Output:
- model: an object containing variables and methods needed to make a prediction

Unsupervised Interface:

model = matLearn_<TYPE>_<NAME>(X, options)

X: n-by-d design matrix
options: additional fields specified by individual function documentation
Output:
- model: an object containing variables and methods needed to make a prediction
- X: a design matrix to project into the reduced-dimension space learned by the model

Looking for an example of a particular machine learning model by name? Check out the "List of files" under each machine learning category below.

Data Description

In the task examples, a number of synthetic and real datasets are used to exhibit strengths and weaknesses of the algorithms. The datasets are described here.

Examples by Task

Regression

List of files

Examples:

Binary Classification

List of files

Additional Input Specifications:

y: elements must one of {-1, 1}

Examples:

Multi-class Classification

List of files

Additional Input Specifications:

Discriminative Models

y: elements must be integer values representing classes

Generative Models

y: elements are disregarded

Examples:

Multilabel Classification

List of files

Examples:

Ordinal Data Classification

List of files

Examples:

Unsupervised Learning

List of files

Examples:

Download

Download link for complete set of files: matLearn2016.zip

Updates

Below are updates to individual pieces of the matLearn package contributed by users:

ml_regression_MLP.m (2017) - fixes a bug in the neural network code.

Authors

The matLearn package contains code from the following sources:

Jennifer (Xin Bei) She organized the previous version of matLearn into a useable form in summer 2015, and added a variety of new methods to the package. A poster prepared by Jennifer is available here.

Geoffrey Roeder merged the existing matLearn library with Mark Schmidt's implementations (dating back to 2004) in summer in 2016, and added a variety of new methods/tasks including unsupervised learning.

Citations

If you use this software in a publication, please cite the version of the work based on the year the package was downloaded using the following information:

G. Roeder, X. She, M. Schmidt, et al. matLearn: machine learning algorithm implementations in Matlab , 2016.
X. She, M. Schmidt, et al. matLearn: machine learning algorithm implementations in Matlab , 2015.

Mark Schmidt > Software > minFunc