The library is written in C/C++. The supported ML task are logistic regression and linear SVM. The supported problem formulation are: s L2-regularized logistic regression, L2-loss and L1-loss linear SVMs. The approach for L1-SVM and L2-SVM is a coordinate descent method. For LR and also L2-SVM, LibLinear implements a trust region Newton method. For multi-class problems, LibLinear implements the one-vs-the-rest strategy and Crammer and Singer method. The SVM learning code from the library is often reused in other open source ML toolkits, including GATE, KNIME, Orange and scikit-learn. The library is very popular in the open source ML community (it is released under the 3-clause BSD license). LibLinear version 2.20 was released on December, 2017.

Strong points

• Designed to solve large-scale linear classification problems.
• Open source, well-maintained and specialized tool with high popularity in open source ML community.

Weak points

• Limited to logistic regression and linear SVM.

Return to Contemt