Semi-Supervised Support Vector Regression (SVR)

Laplacian Embedded Support Vector Regression (Chen et al., 2012)

The picture below shows the decision surface on Two-Moons data set created by LapESVR when the unlabelled is minimally utilized, $\mu = 1$. Note that only the data points near the center are labelled.

When $\mu = 1000$, the model learns the structure of the two moons through the unlabelled data.

Windows Binary

Currently only Windows binary is supported.

1. Set up VC++ environment variables by running vcvars64.bat.
2. To clean the existing binary.

nmake -f Makefile.win clean

3. To build Windows binary.

nmake -f Makefile.win lib

Python Interface

Currently only Python interface is supported.

Example is given in python/LapESVR.ipynb.

References

Chen, L., Tsang, I. W., & Xu, D. (2012). Laplacian embedded regression for scalable manifold regularization. IEEE Transactions on Neural Networks and Learning Systems, 23(6), 902–915. https://doi.org/10.1109/TNNLS.2012.2190420