COMET

Code (MATLAB) and data for the paper "Learning Sparse Metrics, One Feature at a Time", Y. Atzmon, U. Shalit, G. Chechik, JMLR 2015

Project page

Installation Instructions (tested on linux)

1. Download this repository or clone it using git clone https://github.com/chechiklab/COMET

2. Download and compile the Suitesparse (4.4.5+) Cholesky solver:

2.1 open http://faculty.cse.tamu.edu/davis/SuiteSparse/ and download version 4.4.5 or later

2.2 unpack SuiteSparse to a directory, open MATLAB on that directory and call SuiteSparse_install

2.3 cd CHOLMOD/MATLAB, call cholmod_demo to test that the installation was successful

2.4 write down the full path of the location of CHOLMOD/MATLAB and replace it respectively under initpaths.m

Running the examples

To run the examples with the protein dataset, one should download and install LIBSVM matlab package and add it to the matlab path. http://www.csie.ntu.edu.tw/~cjlin/libsvm/#download

Sparse COMET

1. Follow the instructions above to install SuiteSparse (4.4.5+) and LIBSVM.

2. Open example_comet_sparse_training.m and comment-in the dataset name you would like to train upon ('protein' or 'RCV1_4_5K').

3. Execute example_comet_sparse_training.m from MATLAB.

Dense COMET

1. Follow the instructions above to install SuiteSparse (4.4.5+) and LIBSVM.

2. Open example_comet_dense_training.m and comment-in the dataset name you would like to train upon ('protein' or 'RCV1_4_5K').

3. Execute example_comet_dense_training.m from MATLAB.

How to set the hyper-params

There are some technical clarifications need to be made about settings some of the hyper-params. I will document it during march 2016. Send me an email or open an issue on the github repository, if you would like to use this code beforehand and I will assist you.