%
% simpleR: SIMPLE REGRESSION DEMO.
%        Version: 3.1.1
%        Date   : Apr 2018
%
%    This demo shows the training and testing of several state-of-the-art statistical models for regression.
%
%    simpleRegression.m ....... A demo script running all the methods in a single dataset
%    assessment.m ............. A simple function to evaluate classifiers/regression models
%    /vhgpr ................... The folder contains all necessary functions to run both GPR and VHGPR
%    /standard ................ The folder contains all necessary functions to run standard regression models
%
%   --------------------------------------
%   METHODS: Several statistical algorithms are used:
%   --------------------------------------
%
%    * Least squares Linear regression (LR)
%           -- Note that the solution is not regularized
%
%    * Least Absolute Shrinkage and Selection Operator (LASSO).
%           -- This is a Mathworks implementation so you will need the corresponding Matlab toolbox
%           -- We use a 5-fold cross-validation scheme here
%
%    * Elastic Net (ELASTICNET).
%           -- This is a Mathworks implementation so you will need the corresponding Matlab toolbox
%           -- The tradeoff l_1-norm alpha parameter was fixed to 0.5 and could be also crossvalidated
%           -- We use a 5-fold cross-validation scheme here
%
%    * Decision trees (TREE)
%           -- The minimum number of samples to split a node was fixed to 30 and could be also crossvalidated
%           -- The code for doing pruning is commented
%
%    * Bagging trees (BAGTREE)
%           -- The maximum number of trees was set to 200 but could be also crossvalidated
%
%    * Boosting trees (BOOST)
%           -- The maximum number of trees was set to 200 but could be also crossvalidated
%
%    * Neural networks (NN)
%           -- Functions included to automatically train and test standard 1-layer neural
%              networks using the Matlab functions "train" and "sim". The code might not
%              work in newer versions of Matlab, say >2012
%           -- The number of hidden neurons is crossvalidated but no regularization is included
%
%    * Extreme Learning Machines (ELM)
%           -- The standard version of the ELM with random initialization of the weights
%              and pseudoinverse of the output spanning subspace.
%           -- The number of hidden neurons is crossvalidated but no regularization is included
%
%    * Support Vector Regression (SVR)
%           -- Standard support vector implementation for regression and function approximation using the libsvm toolbox.
%           -- Three parameters are adjusted via xval: the regularization term C, the \varepsilon insensitivity
%              tube (tolerated error) and a kernel lengthscale \sigma.
%           -- We include Matlab wrappers for automatic training of the SVR. The
%              wrappers call libsvm compiled functions for training and testing.
%           -- The original source code of libsvm can be obtained from http://www.csie.ntu.edu.tw/~cjlin/libsvm/
%              Please cite the original implementation when appropriate.
%
%           -- We also include our own compilation of the libsvm functions for
%              Linux, Windows and Mac. You are encouraged to use our source and binaries for other
%              platforms in http://www.uv.es/~jordi/soft.htm
%
%             [Smola, 2004] A. J. Smola and B. Schölkopf, “A tutorial on support vector regression,"
%              Statistics and Computing, vol. 14, pp. 199–222, 2004.
%
%    * Kernel Ridge Regression (KRR), aka Least Squares SVM
%           -- Standard least squares regression in kernel feature space.
%           -- Two parameters are adjusted: the regularization term \lambda and an RBF kernel lengthscale \sigma.
%
%    * Relevance Vector Machine (RVM)
%
%           -- We include here the MRVM implementation by Arasanathan Thayananthan (at315@cam.ac.uk)
%              (c) Copyright University of Cambridge
%           -- Please cite the original implementation when appropriate.
%
%              [Thayananthan 2006] Multivariate Relevance Vector Machines for Tracking
%                        Arasanathan Thayananthan et al. (University of Cambridge)
%                        in Proc. 9th European Conference on Computer Vision 2006.
%
%    * Gaussian Process Regression (GPR)
%           -- We consider an anisotropic RBF kernel that has a scale, lengthscale
%              per input feature (band), and a constant noise power parameter as hyperparameters
%           -- The full GP toolbox can be downloaded from http://www.gaussianprocess.org/gpml
%              We include here just two functions "gpr.m" and "minimize.m" in the
%              folder /vhgpr for the sake of convenience.
%           -- Please cite the original implementation when appropriate.
%
%              [Rasmussen 2006] Carl Edward Rasmussen and Christopher K. I. Williams
%                   Gaussian Processes for Machine Learning
%                   The MIT Press, 2006. ISBN 0-262-18253-X.
%
%    * Variational Heteroscedastic Gaussian Process Regression (VHGPR)
%           -- We consider an anisotropic RBF kernel that has a scale, lengthscale
%              per input feature (band), and a input-dependent noise power parameter as hyperparameters
%           -- The original source code can be downloaded from http://www.tsc.uc3m.es/~miguel/
%              Here we include for convenience. If you're interested in VHGPR, please cite:
%
%              [Lázaro-Gredilla, 2011] M. Lázaro-Gredilla and M. K. Titsias, "Variational
%                     heteroscedastic gaussian process regression,"
%                     28th International Conference on Machine Learning, ICML 2011.
%                     Bellevue, WA, USA: ACM, 2011, pp. 841–848.
%
%   --------------------------------------
%   NOTE:
%   --------------------------------------
%
%   All the programs included in this package are intended for illustration
%   purposes and as accompanying software for the paper:
%
%           Miguel Lázaro-Gredilla, Michalis K. Titsias, Jochem Verrelst and
%           Gustavo Camps-Valls. "Retrieval of Biophysical Parameters with
%           Heteroscedastic Gaussian Processes". IEEE Geoscience and Remote
%           Sensing Letters, 2013
%
%   Shall the software is useful for you in other geoscience and remote sensing applications,
%   we would greatly acknowledge citing our paper above. Also, please consider
%   citing these papers for particular methods included herein:
%
%   [KRR, NN]  Nonlinear Statistical Retrieval of Atmospheric Profiles from MetOp-IASI and MTG-IRS Infrared Sounding Data
%     Gustavo Camps-Valls, Jordi Muñoz-Marí, Luis Gómez-Chova, Luis Guanter and Xavier Calbet
%     IEEE Transactions on Geoscience and Remote Sensing, 50(5), 1759 - 1769 2012
%
%   [SVR]  Robust Support Vector Regression for Biophysical Parameter Estimation from Remotely Sensed Images
%     Gustavo Camps-Valls, L. Bruzzone, Jose L. Rojo-Álvarez, Farid Melgani
%     IEEE Geoscience and Remote Sensing Letters, 3(3), 339-343, July 200
%
%   [RVM]  Retrieval of Oceanic Chlorophyll Concentration with Relevance Vector Machines
%     G. Camps-Valls, L. Gomez-Chova, J. Vila-Francés, J. Amorós-López, J. Muñoz-Marí, and J. Calpe-Maravilla
%     Remote Sensing of Environment. 105(1), 23-33, 2006
%
%   [GPR]  Retrieval of Vegetation Biophysical Parameters using Gaussian Processes Techniques
%     J. Verrelst, L. Alonso, G. Camps-Valls, J. Delegido and J. Moreno
%     IEEE Transactions on Geoscience and Remote Sensing, 50(5), 1832 - 1843. 2012
%
%   [GPR/VHGPR]  Retrieval of Biophysical Parameters with Heteroscedastic Gaussian Processes
%     Miguel Lázaro-Gredilla, Michalis K. Titsias, Jochem Verrelst and Gustavo Camps-Valls.
%     IEEE Geoscience and Remote Sensing Letters, 2013
%
%   --------------------------------------
%   Copyright & Disclaimer
%   --------------------------------------
%
%   The programs contained in this package are granted free of charge for
%   research and education purposes only. Scientific results produced using
%   the software provided shall acknowledge the use of this implementation
%   provided by us. If you plan to use it for non-scientific purposes,
%   don't hesitate to contact us. Because the programs are licensed free of
%   charge, there is no warranty for the program, to the extent permitted
%   by applicable law. except when otherwise stated in writing the
%   copyright holders and/or other parties provide the program "as is"
%   without warranty of any kind, either expressed or implied, including,
%   but not limited to, the implied warranties of merchantability and
%   fitness for a particular purpose. the entire risk as to the quality and
%   performance of the program is with you. should the program prove
%   defective, you assume the cost of all necessary servicing, repair or
%   correction. In no event unless required by applicable law or agreed to
%   in writing will any copyright holder, or any other party who may modify
%   and/or redistribute the program, be liable to you for damages,
%   including any general, special, incidental or consequential damages
%   arising out of the use or inability to use the program (including but
%   not limited to loss of data or data being rendered inaccurate or losses
%   sustained by you or third parties or a failure of the program to
%   operate with any other programs), even if such holder or other party
%   has been advised of the possibility of such damages.
%
%   NOTE: This is just a demo providing a default initialization. Training
%   is not at all optimized. Other initializations, optimization techniques,
%   and training strategies may be of course better suited to achieve improved
%   results in this or other problems. We just did it in the standard way for
%   illustration purposes and dissemination of these models.
%
% Copyright (c) 2013 by Gustavo Camps-Valls
% gustavo.camps@uv.es
% http://isp.uv.es/
% http://www.uv.es/gcamps