An efficient and effective algorithm of learning graph for semi-supervised learning. (MATLAB Code)
Before use the code you should compile mtimesx lib, which is inside util/lib/mtimesx/ folder. Please refer to mtimesx. For Mac OS users, you can first use Homebrew to install openblas library, and then run
bias_lib = 'path to libblas.dylib'
mex('-DDEFINEUNIX','-largeArrayDims','mtimesx.c',blas_lib)
After install required library, you should excute main.m in the root folder. After that you can run all matlab files under root folder.
In the example folder, we provide examples with respect to single-thread version PG-Learn, hyperband-parallel version PG-Learn, and several baselines including grid search, random search, MinEnt, AEW and IDML. What's more, we also provide example of running the relatively general parallel framework RndSet_parallel_framework.
Notice that the codes of baselines are kind of messy. These codes are designed for personal usage.
There are six wide-use image datasets under datasets folder. These datasets are benchmark datasets used in our paper. Under each specific benckmark dataset folder, the DatasetName.mat file is the original data. All other *PerTrain subfolders are used specifically by LoadData function. The LoadData function is a personal used function to transform the original data into the data required by algorithms.
Please cite our paper if you use our code in your research.