ConvolutionalLogisticPolicy Class Reference [Auxilliary Tools]

Auxilliary Tool: A class for implementing convolutional logistic policies, which were first used in the IPOMDP domain in Butko and Movellan, 2008 (see Related Publications). More...

#include <ConvolutionalLogisticPolicy.h>

Inherited by MOConvolutionalLogisticPolicy.

List of all members.

Public Member Functions
	ConvolutionalLogisticPolicy (CvSize gridSize)
	Constructor: Create a CLP with space allocated to hold parameters for a grid of size gridSize.
	ConvolutionalLogisticPolicy (ConvolutionalLogisticPolicy *clpToCopy)
	Deep Copy Constructor: Create a CLP that is identical to the one to copy.
virtual	~ConvolutionalLogisticPolicy ()
	Default Destructor.
virtual CvPoint	getFixationPoint (IplImage *currentBelief)
	Pick an appropriate point to fixate for the current belief-state (probability that the object is located at each point).
double	getReward (IplImage *currentBelief)
	Compute the infomax reward associated with this belief state.
void	setPolicy (int policyNumber)
	Tell the CLP what kind of convolution policy to use in the future.
void	setHeuristicPolicyParameters (double softmaxGain, double boxSize)
	Tell the CLP the shape of the convolution kernel to use.
Static Public Attributes
static const int	BOX = 1
static const int	GAUSSIAN = 2
static const int	IMPULSE = 3
static const int	MAX = 4
Friends
std::ostream &	operator<< (std::ostream &ofs, ConvolutionalLogisticPolicy *a)
	Write to a file.
std::istream &	operator>> (std::istream &ifs, ConvolutionalLogisticPolicy *a)
	Read from a file.

---

Detailed Description

Auxilliary Tool: A class for implementing convolutional logistic policies, which were first used in the IPOMDP domain in Butko and Movellan, 2008 (see Related Publications).

Currently, only a few special cases of CLPs are implimented. In the future, we hope to have the most general case coded, along with policy-gradient based optimization methods.

The following policies are implemented:

• BOX: Convolution kernel is a box-filter.
• GAUSSIAN: Convolution kernel is a gaussian filter.
• IMPULSE: Convolution kernel is an impulse response.
• MAX: Convolution kernel is an impulse response with infinite weight.

In the CLP, the belief-image is convolved with some kernel, and then a next fixation is chosen by computing a soft-max over the convolution, and then sampling from the result. The shape of the convolution kernel can be optimized via policy gradient.

Author:

Nicholas Butko

Date:

2010

Version:

0.4

---

Constructor & Destructor Documentation

ConvolutionalLogisticPolicy::ConvolutionalLogisticPolicy

(

CvSize

gridSize )

Constructor: Create a CLP with space allocated to hold parameters for a grid of size gridSize.

The default kernel is a Gaussian envelope with standard-deviation of five grid cells, and a peak value of 300. Use setPolicy() and setHeuristicPolicyParameters() after instantiating the policy to change this. Generally this is done indirectly via the MIPOMDP setPolicy() and setHeuristicPolicyParameters() functions.

ConvolutionalLogisticPolicy::~ConvolutionalLogisticPolicy

(

)

[virtual]

Default Destructor.

Deallocates all memory associated with the CLP.

---

Member Function Documentation

CvPoint ConvolutionalLogisticPolicy::getFixationPoint

(

IplImage *

currentBelief )

[virtual]

Pick an appropriate point to fixate for the current belief-state (probability that the object is located at each point).

In the CLP, the belief-image is convolved with some kernel, and then a next fixation is chosen by computing a soft-max over the convolution, and then sampling from the result.

Parameters:

currentBelief

The current posterior probability estimate that the object we are searching for is located in each respective grid-cell. This belief is maintained and updated by an MIPOMDP object.

Returns:

The grid-cell that should be fixated by the eyes next. Note that in most cases this is a stochastic output, and calling this function repeatedly for different results will result in different return values.

double ConvolutionalLogisticPolicy::getReward

(

IplImage *

currentBelief )

Compute the infomax reward associated with this belief state.

Parameters:

currentBelief

The current posterior probability estimate that the object we are searching for is located in each respective grid-cell. This belief is maintained and updated by an MIPOMDP object.

Returns:

The scalar infomax reward associated with this belief state.

void ConvolutionalLogisticPolicy::setHeuristicPolicyParameters	(	double	softmaxGain,
		double	boxSize
	)

Tell the CLP the shape of the convolution kernel to use.

Parameters:

softmaxGain	Has the following effect for: ConvolutionalLogisticPolicy::BOX - The kernel is a square with integral=softmaxGain ConvolutionalLogisticPolicy::GAUSSIAN - The kernel is a Gaussian with integral=softmaxGain ConvolutionalLogisticPolicy::IMPULSE - The kernel is an impulse response with value softmaxGain ConvolutionalLogisticPolicy::MAX - No effect
boxSize	Has the following effect for: ConvolutionalLogisticPolicy::BOX - The kernel is a boxSize x boxSize grid-cell square. ConvolutionalLogisticPolicy::GAUSSIAN - The kernel is a Gaussian with standard-deviation boxSize grid-cells. ConvolutionalLogisticPolicy::IMPULSE - No effect ConvolutionalLogisticPolicy::MAX - No effect

void ConvolutionalLogisticPolicy::setPolicy

(

int

policyNumber )

Tell the CLP what kind of convolution policy to use in the future.

Parameters:

policyNumber

Must be one of: ConvolutionalLogisticPolicy::BOX ConvolutionalLogisticPolicy::GAUSSIAN ConvolutionalLogisticPolicy::IMPULSE ConvolutionalLogisticPolicy::MAX

---

Member Data Documentation

const int ConvolutionalLogisticPolicy::BOX = 1 [static]

ConvolutionalLogisticPolicy::BOX - Use box-filter for convolution kernel.

const int ConvolutionalLogisticPolicy::GAUSSIAN = 2 [static]

ConvolutionalLogisticPolicy::GAUSSIAN - Use gaussian-envelope for convolution kernel.

const int ConvolutionalLogisticPolicy::IMPULSE = 3 [static]

ConvolutionalLogisticPolicy::IMPULSE - Use impulse response for convolution kernel.

const int ConvolutionalLogisticPolicy::MAX = 4 [static]

ConvolutionalLogisticPolicy::MAX - Use infinite-impulse response for convolution kernel (fixate mode of belief-distribution).

---

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

• /Users/nick/projects/NickThesis/Code/OpenCV/src/ConvolutionalLogisticPolicy.h
• /Users/nick/projects/NickThesis/Code/OpenCV/src/ConvolutionalLogisticPolicy.cpp