gcop::GP Class Reference

#include <gp.h>

List of all members.

Public Member Functions
	GP (int d, int n=0)
virtual	~GP ()
virtual void	Sample ()
bool	Add (const Eigen::VectorXd &x, double f)
void	Train ()
void	Train (const Eigen::MatrixXd &Xs, const Eigen::VectorXd &fs)
double	Predict (const Eigen::VectorXd &x, double *s=0) const
double	SqExp (const Eigen::VectorXd &xa, const Eigen::VectorXd &xb) const
double	LogL (double dll[2]=0)
double	PI (const Eigen::VectorXd &x, double fmin) const
double	OptParams ()
Public Attributes
int	d
	dimension
int	n
	number of data points
Eigen::MatrixXd	Xs
	data points
Eigen::VectorXd	fs
	values
Eigen::MatrixXd	K
Eigen::MatrixXd	Ki
Eigen::MatrixXd	L
Eigen::VectorXd	a
double	l
double	s
double	sigma
bool	cf
	propagate cholesky factor L rather than K^{-1}
bool	eps
	prohibit adding points that are eps-close in L_2 to existing data

---

Constructor & Destructor Documentation

GP::GP	(	int	d,
		int	n = 0
	)

Initialize a GP with dimension d and number of points n

Parameters:

d	dimension
n	number of points (optional)

GP::~GP

(

)

[virtual]

---

Member Function Documentation

bool GP::Add	(	const Eigen::VectorXd &	x,
		double	f
	)

Add a new data point

Parameters:

x	data vector
f	value
true	if OK

References a, cf, d, eps, fs, K, Ki, L, l, n, sigma, SqExp(), and Xs.

double GP::LogL

(

double

dll[2] = 0

)

Loglikelihood

Parameters:

dll	derivative of log-liklihood w.r. to l and s

Returns:

log-likelihood

References a, d, L, l, n, and s.

Referenced by OptParams().

double GP::OptParams

(

)

Optimize GP parameters. This is currently done naively using a grid enumeration over l and s

References l, LogL(), and Train().

double GP::PI	(	const Eigen::VectorXd &	x,
		double	fmin
	)		const

Probability of improvement over a given value fmin x data vector

Parameters:

fmin

baseline

Returns:

probability of improvement

References gcop::ncdf(), Predict(), and s.

double GP::Predict	(	const Eigen::VectorXd &	x,
		double *	s = 0
	)		const

Predict value at point x

Parameters:

x	point
s	pointer to predicted covariance (optional)

Returns:

predicted mean

References a, cf, Ki, L, sigma, and SqExp().

Referenced by PI().

void GP::Sample

(

)

[virtual]

Generate random points

References fs, and Xs.

double GP::SqExp	(	const Eigen::VectorXd &	xa,
		const Eigen::VectorXd &	xb
	)		const

Square exponential kernel

Parameters:

xa	first point
xb	second point

Returns:

correlation

References d, l, and s.

Referenced by Add(), Predict(), and Train().

void GP::Train

(

)

Train using current data

References a, cf, K, Ki, L, n, sigma, and SqExp().

Referenced by OptParams().

void gcop::GP::Train	(	const Eigen::MatrixXd &	Xs,
		const Eigen::VectorXd &	fs
	)

Train GP using a given dataset (xs, fs)

Parameters:

d-n	matrix of data vectors
n-vector	of values

---

Member Data Documentation

Eigen::VectorXd gcop::GP::a

Referenced by Add(), LogL(), Predict(), and Train().

bool gcop::GP::cf

propagate cholesky factor L rather than K^{-1}

Referenced by Add(), Predict(), and Train().

int gcop::GP::d

dimension

Referenced by Add(), LogL(), and SqExp().

bool gcop::GP::eps

prohibit adding points that are eps-close in L_2 to existing data

Referenced by Add().

Eigen::VectorXd gcop::GP::fs

values

Referenced by Add(), and Sample().

Eigen::MatrixXd gcop::GP::K

Referenced by Add(), and Train().

Eigen::MatrixXd gcop::GP::Ki

Referenced by Add(), Predict(), and Train().

Eigen::MatrixXd gcop::GP::L

Referenced by Add(), LogL(), Predict(), and Train().

double gcop::GP::l

Referenced by Add(), LogL(), OptParams(), and SqExp().

int gcop::GP::n

number of data points

Referenced by Add(), LogL(), and Train().

double gcop::GP::s

Referenced by LogL(), PI(), and SqExp().

double gcop::GP::sigma

Referenced by Add(), Predict(), and Train().

Eigen::MatrixXd gcop::GP::Xs

data points

Referenced by Add(), and Sample().

---

The documentation for this class was generated from the following files:

• lib/est/gp.h
• lib/est/gp.cc