########################################################################

	Sparse Gaussian Process Regression v1.0
	--- post-backfitting matching pursuit  
	--- pre-backfitting matching pursuit  
	--- information gain approach
	--- baseline random selection
	
	NOTE: model parameters should be specified by user. 
	
	SOURCE CODE: http://www.gatsby.ucl.ac.uk/~chuwei/code/sgpr.tar

Wei CHU (C)Copyright 2005-2006 at Gatsby Unit.

########################################################################

1. Installation
	a) download sgpr.tar to your machine (Linux is assumed)
	b) tar -xvf sgpr.tar
	c) cd sgpr
	d) the program of post-backfitting matching pursuit is generated by
	   gcc -o sgpr *.c -lg2c -lm -O2 -Wall -D _GPR_KEERTHI
	e) the program of pre-backfitting matching pursuit is generated by
	   gcc -o sgpr *.c -lg2c -lm -O2 -Wall -D _GPR_SMOLA
	f) the program of information gain approach is generated by
	   gcc -o sgpr *.c -lg2c -lm -O2 -Wall -D _GPR_SEEGER
	g) the program of baseline random selection is generated by
	   gcc -o sgpr *.c -lg2c -lm -O2 -Wall 
	h) the program using ARD kernel is generated by
	   gcc -o sgpr *.c -lg2c -lm -O2 -Wall -D _GPRARD
	i) the program using normalized inputs is generated by
	   gcc -o sgpr *.c -lg2c -lm -O2 -Wall -D _GPRNORMALIZEINPUT
	j) the program using normalized targets is generated by
	   gcc -o sgpr *.c -lg2c -lm -O2 -Wall -D _GPRNORMALIZETARGET
 
2. Data Format
	a) a space delimited plain text file with a new-line character 
	 at the end of each line
	b) each line contains one sample.
	c) On each line, the last element is the target, a real value
	e) training data saved in a text file, say "mytask_train.1".
	f) the data for test with/without targers saved in "mytask_test.1".
	g) the test targets are saved in "mytask_targets.1" if available.
	h) "mytask" could be any string rather than containing "train" and "test" 
	followed by "_train" and "_test".
	i) see the example files under the folder:
	  sinctoy_train.dat and sinctoy_test.dat 	
	
3. Input and Output
	a) prepare your data files
		"mytask_train.1"
		"mytask_test.1"
		"mytask_targets.1" (optional)
	b) run the program (specify your model parameters as well !!)
		./sgpr mytask_train.1
	c) if _GPRARD has been defined, the file "mytask_train.1.ard" 
	could be supplied to specify the values of ARD parameters
	d) output files
		1) "mytask_test.1.std" to save the std of predictive
		 distribution of test data.
		2) "mytask_test.1.mean" to save the mean of predictive
		 distribution of test data.
		3) a log file to save results for batch tasks. Each line contains: 
		ase nmse nlpd aae rmse qgap time.
	
4. Options
	a) for help information
		./sgpr
	b) specify initial noise variance, say 0.4
		./sgpr -S 0.4 mytask_train.1
	c) specify initial kernel parameter, say 0.01
		./sgpr -K 0.01 mytask_train.1
	d) specify the power level, say 3.0
		./gpor -O 3 mytask_train.1
	e) specify the number of basis functions, say 1000
		./sgpr -D 1000 mytask_train.1

5. Reference
	"A matching pursuit approach to sparse Gaussian process regression",
	S. S. Keerthi and W. Chu 
	NIPS-18 2005
	http://www.gatsby.ucl.ac.uk/~chuwei/paper/sgpr.ps

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

This program is free software; you can redistribute it and/or
modify it under the terms of the GNU General Public License
as published by the Free Software Foundation; either version 2
of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU General Public License for more details.

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~