Kernel regression for signals over graphs
(c) Arun Venkitaraman, 2018.

This package contains code implementing the concepts presented in the 
following paper: 

[1] Arun Venkitaraman, Saikat Chatterjee, and Peter Händel, 
“Kernel regression for signals over graphs" [Online]: https://arxiv.org/abs/1706.02191

Since most of the data such as the Swedish temperature data, and the computation of the geodesic distance for ETEX data were computed by us during the evaluation of the experiments, since they are not available elsewhere to our knowledge, we ask you to Kindly cite our article when you use the data for your purpose. Thanks!


The package has the following files:
main_KRG: Main file computes LR, LRG, KR, and KRG with specified graph
return_KRG_xvalparams.m: Returns the optimal regularisation parameters for LR, LRG, KR, and KRG given the input data
KerRegGraph_fast.m: Code to compute the optimal regression coefficient \Psi
matrix_max_loc.m: computes the location of the smallest entry in an input matrix
get_dataset.m: returns relevant data depending on choice of dataset
Datasets:


For reproducing the synthesised graph examples, please install GSPbox toolbox available here:
https://epfl-lts2.github.io/gspbox-html/doc/base/

     Swedish temperature data:
      City45data.mat: GPS coordinates o all cities, and matrix containing geodesic distance In   kilometres
      smhi_temp_17.mat: Temperature measurements of three months in 2017 
      smhi17_partition.mat: the data indices of the partition of total data into test and training set used in experiments

ETEX atmospheric tracer diffusion data:
  etex_1.mat, etex_2.mat: Data for two experiments of ETEX
  etex_partition.mat:the data indices of the partition of total data into test and training set used in experiments
A_etex: geodesic distance between all the 183 cities involved in the experiment

EEG data
 eegdata_S002R01.mat: the 64 channel EEG data used in experiment
 EEG_partition.mat: the data indices of the partition of total data into test and training set used in experiments

Cerebellum fMRI data:
A_cerebellum.mat: adjacency matrix of the different voxels
signal_Set_cerebellum.mat: the corresponding graph signals
Cere_partition.mat: the data indices of the partition of total data into test and training set used in experiments


Good Luck!     

Arun Venkitaraman
March 2017

