Multi-Class Learning: From Theory to Algorithm

This repository provides the code used to run the experiments of the paper "Multi-Class Learning: From Theory to Algorithm" (http://papers.nips.cc/paper/7431-multi-class-learning-from-theory-to-algorithm), which has been published in NeurIPS 2018.

Usage of source code

Code used in experiments locates in ./code

Enviroment

We do experiments based on following softwares:

1. Python 2.7
2. MATLAB R2017b
3. DOGMA toolbox from http://dogma.sourceforge.net/
4. SHOGUN-6.1.3 from https://github.com/shogun-toolbox/shogun
5. LIBSVM Tools from https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/
6. sklearn for python

Data sets

1. plant, psortPos, psortNeg and nonpl from http://www.raetschlab.org/suppl/protsubloc
2. others from https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/

Steps

1. Download data sets and move dataName.phylpro.mat, label_dataName.mat and dataName.scale to code/data/
2. Create Gaussian kernels: change variable file_list in Test_Gaussian_Kernel.m and run
3. Run following methods

• SMSD-MKL: change variables of data_sets in Test_SMSD_MKL.m and run
• Conv-MKL: change variables of data_sets in Test_Conv_MKL.m and run
• LMC: change variables of data_sets in Test_LMC.py and run
• OneVsOne: change variables of data_sets in Test_OneVsOne.m and run
• OneVsRest: change variables of data_sets in Test_OneVsOne.m and run
• GMNP: change variables of data_sets in Test_GMNP.py and run
• l1 MC-MKL: change variables of data_sets in Test_MC_MKL_1.py and run
• l2 MC-MKL: change variables of data_sets in Test_MC_MKL_2.py and run
• UFO-MKL: change variables of data_sets in Test_UFO_MKL.m and run