USMO (Unlabeled data in Sequential Minimal Optimization)

Matlab code for paper Efficient Training for Positive Unlabeled Learning

How to run the code

  1. Install dependencies. Download LIBSVM, extract the archive into the main directory of USMO and finally compile the Matlab version of LIBSVM (use the make.m file in the uncompressed folder).

  2. Use demo1.m and demo2.m as examples to call USMO routine.

Demo 1

Classification of MNIST dataset (after applying PCA to visualize data)

Linear kernel Polynomial kernel Gaussian kernel

Legend. Red and black points are positive and negative samples, respectively. Triangles are used to identify labeled samples.

Licensing

The code is provided “as-is” under the terms of General Public License. See LICENSE for full details.