LIBSVM

Description

LIBSVM is a comprehensive and easy-to-use software library designed for Support Vector Machines (SVM). It provides implementations for classification (C-SVC, nu-SVC), regression (epsilon-SVR, nu-SVR), and distribution estimation (one-class SVM). The library is engineered to be accessible to users from diverse fields who wish to leverage SVM as a powerful tool.

Key features of LIBSVM include efficient multi-class classification capabilities and support for various SVM formulations. It offers a range of kernel functions, including linear, polynomial, radial basis function (RBF), and sigmoid, with options for precomputed kernel matrices. For handling imbalanced datasets, LIBSVM supports weighted SVM. The library also provides cross-validation for model selection and probability estimates for classification and regression tasks.

LIBSVM is designed with a simple interface, allowing users to easily integrate it into their own programs. It is available with both C++ and Java source code, and includes interfaces for numerous programming languages and environments such as Python, R, MATLAB, Perl, Ruby, Weka, Common LISP, CLISP, Haskell, OCaml, LabVIEW, and PHP. Additionally, C# .NET code and CUDA extensions are available, and it is integrated into data mining environments like RapidMiner and PCP.

The package includes tools for data scaling and automatic model selection, such as the `grid.py` script which generates contour plots of cross-validation accuracy, aiding in parameter optimization. The project has a history of successful applications, with its developers winning several challenges, including the IJCNN 2001 Challenge and the WCCI 2008 Causation and Prediction challenge. The latest version, 3.37, was released in December 2025, with ongoing updates and bug fixes.

LIBSVM aims to democratize the use of SVM by providing a robust, efficient, and well-documented library. Its extensive language support and integration capabilities make it a versatile choice for researchers and developers working on machine learning tasks. The project also maintains a page for LIBSVM datasets and offers practical guides for beginners.

LIBSVM's Core Features

• Supports C-SVC, nu-SVC, one-class SVM, epsilon-SVR, and nu-SVR

• Efficient multi-class classification

• Various kernel functions: linear, polynomial, RBF, sigmoid

• Weighted SVM for unbalanced data

• Cross-validation for model selection

• Probability estimates

• Includes data scaling tools

• Automatic model selection with parameter optimization

• C++ and Java source code available

• Interfaces for Python, R, MATLAB, Perl, Ruby, Weka, LISP, Haskell, OCaml, LabVIEW, PHP, C#, CUDA, Node.js, Javascript, Scilab, Nimrod, Julia, Android, GO, Gretl

• Integrated into data mining environments like RapidMiner

• Installable via package managers like pip and vcpkg

Getting Started with LIBSVM

1. 1

   Download LIBSVM: Obtain the latest release from the official website or GitHub repository.

2. 2

   Install Interface: Install the appropriate language interface (e.g., `pip install -U libsvm-official` for Python).

3. 3

   Prepare Data: Format your training and testing data, potentially using provided scaling tools.

4. 4

   Configure Parameters: Select SVM type, kernel, and other parameters using command-line options or API calls.

5. 5

   Train Model: Use the LIBSVM library or command-line tool to train your SVM model.

6. 6

   Evaluate Model: Perform cross-validation or use probability estimates to assess model performance.

7. 7

   Integrate: Link LIBSVM into your own programs or data mining workflows.

LIBSVM's Use Cases

• Classification Tasks
• Regression Analysis
• Distribution Estimation
• Parameter Optimization
• Data Mining
• Machine Learning Research
• Load Prediction
• Feature Selection

FAQ from LIBSVM

• What is LIBSVM?

  LIBSVM is an integrated software package for support vector classification, regression, and distribution estimation, designed for ease of use and efficiency.

• What types of SVM does LIBSVM support?

  LIBSVM supports C-SVC, nu-SVC, one-class SVM, epsilon-SVR, and nu-SVR.

• Can LIBSVM handle multi-class classification?

  Yes, LIBSVM implements efficient multi-class classification.

• What kernel functions are available in LIBSVM?

  LIBSVM offers linear, polynomial, radial basis function (RBF), and sigmoid kernels, along with support for precomputed kernel matrices.

• How can I install LIBSVM for Python?

  You can install the Python interface using pip: `pip install -U libsvm-official`.

• Does LIBSVM support probability estimates?

  Yes, LIBSVM supports probability estimates for classification and regression tasks.

• Is LIBSVM suitable for beginners?

  Yes, LIBSVM provides a simple interface, a practical guide for beginners, and an easy script (`easy.py`) for automatic data scaling and parameter selection.

• Can LIBSVM be used with GPU acceleration?

  Yes, a CUDA extension is available for LIBSVM, enabling GPU acceleration.

• Where can I find datasets for LIBSVM?

  LIBSVM provides a dedicated page for LIBSVM datasets, offering problems in LIBSVM format.

• What is the latest version of LIBSVM?

  The latest release is Version 3.37, released on December 29, 2025.

• How can I cite LIBSVM in my research?

  You can cite LIBSVM using the ACM Transactions on Intelligent Systems and Technology paper: C.-C. Chang and C.-J. Lin. LIBSVM : a library for support vector machines. ACM Transactions on Intelligent Systems and Technology, 2:27:1--27:27, 2011.

• Does LIBSVM offer a graphical interface?

  Yes, there is a simple applet demonstrating SVM classification and regression.