***************************************************************

     Example Codes for State Space Methods for Efficient 
         Inference in Student-t Process Regression

***************************************************************

INTRODUCTION

  This package acts as a supplementary code package for the 
  following paper:

  @inproceedings{Solin+Sarkka:2015,
     author = {Solin, Arno and S\"arkk\"a, Simo},
      title = {State space methods for efficient inference in 
               {S}tudent-$t$ process regression},
  booktitle = {Proceedings of the 18th International Conference
               on Artificial Intelligence and Statistics},
     series = {{JMLR} Workshop and Conference Proceedings},
       year = {2015},
     volume = {38},
      pages = {885--893}
  }

  Available online at: 
    http://jmlr.org/proceedings/papers/v38/solin15.html

  If you find the methods useful in your research or work, 
  please cite the paper.


DESCRIPTION:

  This code package features the methods for conducting Student-t
  process regression by converting the TP regression problem
  to a state space form stochastic differential equation, and 
  doing the inference recursively using a Student-t filter and
  smoother.
  
  For comparison we have also included a naive implementation of
  of solving the TP regression problem without the filter/smoother,
  and also the corresponding methods for GP inference.
  

PACKAGE CONTENTS:

* METHODS

  The proof-of-concept implementations as Mathworkds Matlab code 
  are included for the following methods `tf_solve.m' (state space 
  TP regression), `tp_solve.m' (naive TP regression), `gf_solve.m' 
  (state space GP regression), and `gp_solve.m' (naive GP 
  regression).

  Additionally an example `example.m' of how to use the methods
  and some house-keeping codes are also included.

* GPS DATA

  The data that was used in the GPS interpolation example in the
  paper. The columns are the following: time (seconds), location
  x-coordinate, location y-coordinate. The file is in the CSV 
  format.


NOTES:

* The codes were tested under Matlab 2014b they should to
  most part be compatible with Octave as well.

* Codes for conversions between kernel representations of
  covariance functions and their corresponding state space
  SDEs can be found, for example, in the GPstuff toolbox now.
  

VERSION:

  This package was updated: 2015-05-04

COPYRIGHT:
  
  (c) Arno Solin and Simo Särkkä, 2015

      The codes from the SGP toolbox are copyrighted by 
      Arno Solin, Jukka Koskenranta and Simo Särkkä.


LICENSE:

  This software is distributed under the GNU General Public
  License (version 3 or later); please refer to the file
  license.txt, included with the software, for details.  



