Incremental and Decremental Learning Extension of LIBLINEAR

This extension of LIBLINEAR supports incremental and decremental learning. If some data are added or removed, it applies a warm start strategy to efficiently updates the previously trained model. Technical details are in the following paper.

C.-H. Tsai, C.-Y. Lin, and C.-J. Lin. Incremental and decremental training for linear classification. KDD 2014.

If you successfully used this code for your applications, please let us know. We are interested in how it's being used.

How to Run this LIBLINEAR Extension

Please download the zip file. See README for building the package and README.incdec for details of running this extension.

The usage is the same as LIBLINEAR except a new option "-i." Specify "-i model" to load a previously computed model for quickly training a slightly modified data set. Example:

> train -s 0 heart_scale.set1
> train -s 0 -i heart_scale.set1.model heart_scale.set2

MATLAB/Octave interface is now supported. Please check matlab/README.incdec

Python interface is now supported. Please check python/README.incdec

Supplementary materials

Supplementary materials includes the derivation not shown in the paper and more results of experiments.

Experiment code

You can use this code to reproduce the experiments in both the paper and supplementary materials.

Please contact Chih-Jen Lin for any question.