de.jstacs.models.hmm.states.emissions.discrete
Class AbstractConditionalDiscreteEmission

java.lang.Object
  de.jstacs.models.hmm.states.emissions.discrete.AbstractConditionalDiscreteEmission

All Implemented Interfaces:

DifferentiableEmission, Emission, SamplingEmission, SamplingComponent, SamplingFromStatistic, Storable, Cloneable

Direct Known Subclasses:

DiscreteEmission, ReferenceSequenceDiscreteEmission

public abstract class AbstractConditionalDiscreteEmission
extends Object
implements SamplingEmission, DifferentiableEmission

The abstract super class of discrete emissions.

Author:

Jens Keilwagen, Michael Scharfe, Jan Grau

Field Summary

protected AlphabetContainer	con The alphabet of the emissions
protected int[]	counter The counter for the sampling steps of each sampling.
protected double[]	ess The equivalent sample sizes for each condition
protected double[][]	grad The array for storing the gradients for each parameter
protected double[][]	hyperParams The hyper-parameters for the prior on the parameters
protected double[]	logNorm The log-normalization constants for each condition
protected int	offset The offset of the parameter indexes
protected double[][]	params The parameters of the emission
protected File[]	paramsFile The files for saving the parameters during the sampling.
protected double[][]	probs The parameters transformed to probabilites
protected BufferedReader	reader The reader for the paramsFile after a sampling.
protected int	samplingIndex The index of the current sampling.
protected double[][]	statistic The array for storing the statistics for each parameter
protected BufferedWriter	writer The writer for the paramsFile in a sampling.

Constructor Summary

protected	AbstractConditionalDiscreteEmission(AlphabetContainer con, double[][] hyperParams) This is a simple constructor for a AbstractConditionalDiscreteEmission defining the individual hyper parameters.
protected	AbstractConditionalDiscreteEmission(AlphabetContainer con, double[][] hyperParams, double[][] initHyperParams) This constructor creates a AbstractConditionalDiscreteEmission defining the individual hyper parameters for the prior used during training and initialization.
protected	AbstractConditionalDiscreteEmission(AlphabetContainer con, int numberOfConditions, double ess) This is a simple constructor for a AbstractConditionalDiscreteEmission based on the equivalent sample size.
protected	AbstractConditionalDiscreteEmission(StringBuffer xml) Creates a AbstractConditionalDiscreteEmission from its XML representation.

Method Summary

void	acceptParameters() This methods accepts the drawn parameters.
void	addGradientOfLogPriorTerm(double[] gradient, int offset) This method computes the gradient of Emission.getLogPriorTerm() for each parameter of this model.
void	addToStatistic(boolean forward, int startPos, int endPos, double weight, Sequence seq) This method adds the weight to the internal sufficient statistic.
protected void	appendFurtherInformation(StringBuffer xml) This method appends further information to the XML representation.
AbstractConditionalDiscreteEmission	clone()
void	drawParametersFromStatistic() This method draws the parameters using a sufficient statistic representing a posteriori density.
void	estimateFromStatistic() This method estimates the parameters from the internal sufficient statistic.
void	extendSampling(int start, boolean append) This method allows to extend a sampling.
protected void	extractFurtherInformation(StringBuffer xml) This method extracts further information from the XML representation.
void	fillCurrentParameter(double[] params) Fills the current parameters in the global code>params array using the internal offset.
void	fillSamplingGroups(int parameterOffset, LinkedList<int[]> list) Adds the groups of indexes of those parameters of this emission that should be sampled together in one step of a grouped sampling procedure, each as an int[], into list.
protected void	finalize()
protected void	fromXML(StringBuffer xml) This method is internally used by the constructor AbstractConditionalDiscreteEmission(StringBuffer).
AlphabetContainer	getAlphabetContainer() This method returns the AlphabetContainer of this emission.
protected abstract int	getConditionIndex(boolean forward, int seqPos, Sequence seq) This method returns an index encoding the condition.
protected static double[][]	getHyperParams(double ess, int numConditions, int numEmissions) Returns the hyper-parameters for all parameters and a given ess.
double	getLogGammaScoreFromStatistic() This method calculates a score for the current statistics, which is independent from the current parameters In general the gamma-score is a product of gamma-functions parameterized with the current statistics
double	getLogPosteriorFromStatistic() This method calculates the a-posteriori probability for the current statistics
double	getLogPriorTerm() Returns a value that is proportional to the log of the prior.
double	getLogProbAndPartialDerivationFor(boolean forward, int startPos, int endPos, IntList indices, DoubleList partDer, Sequence seq) Returns the logarithmic score for a Sequence beginning at position start in the Sequence and fills lists with the indices and the partial derivations.
double	getLogProbFor(boolean forward, int startPos, int endPos, Sequence seq) This method computes the logarithm of the likelihood.
String	getNodeLabel(double weight, String name, NumberFormat nf) Returns the graphviz label of the node containing this emission.
String	getNodeShape(boolean forward) Returns the graphviz string for the shape of the node.
int	getNumberOfParameters() Returns the number of parameters of this emission.
int	getSizeOfEventSpace() Returns the size of the event space, i.e., the number of possible outcomes, for the random variables of this emission
void	initForSampling(int starts) This method initializes the instance for the sampling.
void	initializeFunctionRandomly() This method initializes the emission randomly.
boolean	isInSamplingMode() This method returns true if the object is currently used in a sampling, otherwise false.
boolean	parseNextParameterSet() This method allows the user to parse the next set of parameters (from a file).
boolean	parseParameterSet(int start, int n) This method allows the user to parse the set of parameters with index n of a certain sampling (from a file).
protected void	precompute() This method precomputes some normalization constant and probabilities.
void	resetStatistic() This method resets the internal sufficient statistic.
void	samplingStopped() This method is the opposite of the method SamplingComponent.extendSampling(int, boolean).
void	setLinear(boolean linear) If set to true, the probabilities are mapped to colors by directly, otherwise a logistic mapping is used to emphasize deviations from the uniform distribution.
void	setParameter(double[] params, int offset) This method sets the internal parameters using the given global parameter array, the global offset of the HMM and the internal offset.
int	setParameterOffset(int offset) This method sets the internal parameter offset and returns the new parameter offset for further use.
void	setShape(String shape) Sets the graphviz shape of the node that uses this emission to some non-standard value (standard is "house").
abstract String	toString()
StringBuffer	toXML() This method returns an XML representation as StringBuffer of an instance of the implementing class.

Methods inherited from class java.lang.Object

equals, getClass, hashCode, notify, notifyAll, wait, wait, wait

Field Detail

paramsFile

protected File[] paramsFile

The files for saving the parameters during the sampling.

counter

protected int[] counter

The counter for the sampling steps of each sampling.

samplingIndex

protected int samplingIndex

The index of the current sampling.

writer

protected BufferedWriter writer

The writer for the paramsFile in a sampling.

reader

protected BufferedReader reader

The reader for the paramsFile after a sampling.

offset

protected int offset

The offset of the parameter indexes

con

protected AlphabetContainer con

The alphabet of the emissions

params

protected double[][] params

The parameters of the emission

probs

protected double[][] probs

The parameters transformed to probabilites

hyperParams

protected double[][] hyperParams

The hyper-parameters for the prior on the parameters

statistic

protected double[][] statistic

The array for storing the statistics for each parameter

grad

protected double[][] grad

The array for storing the gradients for each parameter

logNorm

protected double[] logNorm

The log-normalization constants for each condition

ess

protected double[] ess

The equivalent sample sizes for each condition

Constructor Detail

AbstractConditionalDiscreteEmission

protected AbstractConditionalDiscreteEmission(AlphabetContainer con,
                                              int numberOfConditions,
                                              double ess)

This is a simple constructor for a AbstractConditionalDiscreteEmission based on the equivalent sample size.

Parameters:

con - the AlphabetContainer of this emission

numberOfConditions - the number of conditions

ess - the equivalent sample size (ess) of this emission that is equally distributed over all parameters

See Also:

AbstractConditionalDiscreteEmission(AlphabetContainer, double[][])

AbstractConditionalDiscreteEmission

protected AbstractConditionalDiscreteEmission(AlphabetContainer con,
                                              double[][] hyperParams)

This is a simple constructor for a AbstractConditionalDiscreteEmission defining the individual hyper parameters.

Parameters:

con - the AlphabetContainer of this emission

hyperParams - the individual hyper parameters for each parameter

See Also:

AbstractConditionalDiscreteEmission(AlphabetContainer, double[][])

AbstractConditionalDiscreteEmission

protected AbstractConditionalDiscreteEmission(AlphabetContainer con,
                                              double[][] hyperParams,
                                              double[][] initHyperParams)

This constructor creates a AbstractConditionalDiscreteEmission defining the individual hyper parameters for the prior used during training and initialization.

Parameters:

con - the AlphabetContainer of this emission

hyperParams - the individual hyper parameters for each parameter (used during training)

initHyperParams - the individual hyper parameters for each parameter used in initializeFunctionRandomly()

AbstractConditionalDiscreteEmission

protected AbstractConditionalDiscreteEmission(StringBuffer xml)
                                       throws NonParsableException

Creates a AbstractConditionalDiscreteEmission from its XML representation.

Parameters:

xml - the XML representation.

Throws:

NonParsableException - if the XML representation could not be parsed

Method Detail

getHyperParams

protected static double[][] getHyperParams(double ess,
                                           int numConditions,
                                           int numEmissions)

Returns the hyper-parameters for all parameters and a given ess. The equivalent sample size is distributed evenly across all parameters

Parameters:

ess - the equivalent sample size

numConditions - the number of conditions

numEmissions - the number of emissions, assumed to be equal for all conditions

Returns:

hyper-parameters for all parameters

clone

public AbstractConditionalDiscreteEmission clone()
                                          throws CloneNotSupportedException

Overrides:

clone in class Object

Throws:

CloneNotSupportedException

setShape

public void setShape(String shape)

Sets the graphviz shape of the node that uses this emission to some non-standard value (standard is "house").

Parameters:

shape - the shape of the node

addGradientOfLogPriorTerm

public void addGradientOfLogPriorTerm(double[] gradient,
                                      int offset)

Description copied from interface: DifferentiableEmission

This method computes the gradient of Emission.getLogPriorTerm() for each parameter of this model. The results are added to the array grad beginning at index (offset + internal offset).

Specified by:

addGradientOfLogPriorTerm in interface DifferentiableEmission

Parameters:

gradient - the array of gradients

offset - the start index of the HMM in the grad array, where the partial derivations for the parameters of the HMM shall be entered

See Also:

Emission.getLogPriorTerm(), DifferentiableEmission.setParameterOffset(int)

getLogPriorTerm

public double getLogPriorTerm()

Description copied from interface: Emission

Returns a value that is proportional to the log of the prior. For maximum likelihood (ML) 0 should be returned.

Specified by:

getLogPriorTerm in interface Emission

Returns:

a value that is proportional to the log of the prior

See Also:

Model.getLogPriorTerm()

getLogProbAndPartialDerivationFor

public double getLogProbAndPartialDerivationFor(boolean forward,
                                                int startPos,
                                                int endPos,
                                                IntList indices,
                                                DoubleList partDer,
                                                Sequence seq)
                                         throws OperationNotSupportedException

Description copied from interface: DifferentiableEmission

Returns the logarithmic score for a Sequence beginning at position start in the Sequence and fills lists with the indices and the partial derivations.

Specified by:

getLogProbAndPartialDerivationFor in interface DifferentiableEmission

Parameters:

forward - a switch whether to use the forward or the reverse complementary strand of the sequence

startPos - the start position in the Sequence

endPos - the end position in the Sequence

indices - an IntList of indices, after method invocation the list should contain the indices i where is not zero

partDer - a DoubleList of partial derivations, after method invocation the list should contain the corresponding that are not zero

seq - the Sequence

Returns:

the logarithmic score for the Sequence

Throws:

OperationNotSupportedException - if forward==false and the reverse complement of the sequence can not be computed

getLogProbFor

public double getLogProbFor(boolean forward,
                            int startPos,
                            int endPos,
                            Sequence seq)
                     throws OperationNotSupportedException

Description copied from interface: Emission

This method computes the logarithm of the likelihood.

Specified by:

getLogProbFor in interface Emission

Parameters:

forward - whether to use the forward or the reverse strand

startPos - the start position

endPos - the end position

seq - the sequence

Returns:

the logarithm of the probability

Throws:

OperationNotSupportedException - if forward=false and the reverse complement of the sequence seq is not defined

initializeFunctionRandomly

public void initializeFunctionRandomly()

Description copied from interface: Emission

This method initializes the emission randomly.

Specified by:

initializeFunctionRandomly in interface Emission

precompute

protected void precompute()

This method precomputes some normalization constant and probabilities.

See Also:

logNorm, probs

toXML

public StringBuffer toXML()

Description copied from interface: Storable

This method returns an XML representation as StringBuffer of an instance of the implementing class.

Specified by:

toXML in interface Storable

Returns:

the XML representation

appendFurtherInformation

protected void appendFurtherInformation(StringBuffer xml)

This method appends further information to the XML representation. It allows subclasses to save further parameters that are not defined in the superclass.

Parameters:

xml - the XML representation

fromXML

protected void fromXML(StringBuffer xml)
                throws NonParsableException

This method is internally used by the constructor AbstractConditionalDiscreteEmission(StringBuffer).

Parameters:

xml - the StringBuffer containing the xml representation of an instance

Throws:

NonParsableException - if the StringBuffer is not parsable

See Also:

AbstractConditionalDiscreteEmission(StringBuffer)

extractFurtherInformation

protected void extractFurtherInformation(StringBuffer xml)
                                  throws NonParsableException

This method extracts further information from the XML representation. It allows subclasses to cast further parameters that are not defined in the superclass.

Parameters:

xml - the XML representation

Throws:

NonParsableException - if the information could not be reconstructed out of the StringBuffer xml

addToStatistic

public void addToStatistic(boolean forward,
                           int startPos,
                           int endPos,
                           double weight,
                           Sequence seq)
                    throws OperationNotSupportedException

Description copied from interface: Emission

This method adds the weight to the internal sufficient statistic.

Specified by:

addToStatistic in interface Emission

Parameters:

forward - whether to use the forward or the reverse strand

startPos - the start position

endPos - the end position

weight - the weight of the sequence

seq - the sequence

Throws:

OperationNotSupportedException - if forward=false and the reverse complement of the sequence seq is not defined

getConditionIndex

protected abstract int getConditionIndex(boolean forward,
                                         int seqPos,
                                         Sequence seq)

This method returns an index encoding the condition.

Parameters:

forward - a switch to decide whether to use the forward or the reverse complementary strand (e.g. for DNA sequences)

seqPos - the position in the sequence seq

seq - the sequence

Returns:

the index encoding the condition

estimateFromStatistic

public void estimateFromStatistic()

Description copied from interface: Emission

This method estimates the parameters from the internal sufficient statistic.

Specified by:

estimateFromStatistic in interface Emission

resetStatistic

public void resetStatistic()

Description copied from interface: Emission

This method resets the internal sufficient statistic.

Specified by:

resetStatistic in interface Emission

toString

public abstract String toString()

Overrides:

toString in class Object

setParameter

public void setParameter(double[] params,
                         int offset)

Description copied from interface: DifferentiableEmission

This method sets the internal parameters using the given global parameter array, the global offset of the HMM and the internal offset.

Specified by:

setParameter in interface DifferentiableEmission

Parameters:

params - the global parameter array of the classifier

offset - the offset of the HMM

See Also:

DifferentiableEmission.setParameterOffset(int)

getAlphabetContainer

public AlphabetContainer getAlphabetContainer()

Description copied from interface: Emission

This method returns the AlphabetContainer of this emission.

Specified by:

getAlphabetContainer in interface Emission

Returns:

the AlphabetContainer of this emission

fillCurrentParameter

public void fillCurrentParameter(double[] params)

Description copied from interface: DifferentiableEmission

Fills the current parameters in the global code>params array using the internal offset.

Specified by:

fillCurrentParameter in interface DifferentiableEmission

Parameters:

params - the global parameter array of the HMM

See Also:

DifferentiableEmission.setParameterOffset(int)

setParameterOffset

public int setParameterOffset(int offset)

Description copied from interface: DifferentiableEmission

This method sets the internal parameter offset and returns the new parameter offset for further use.

Specified by:

setParameterOffset in interface DifferentiableEmission

Parameters:

offset - the offset to be set

Returns:

the new parameter offset

drawParametersFromStatistic

public void drawParametersFromStatistic()
                                 throws Exception

Description copied from interface: SamplingFromStatistic

This method draws the parameters using a sufficient statistic representing a posteriori density. It is recommended to write the parameters to a specific file using SamplingComponent.acceptParameters() so that they can later be parsed using the methods of the interface.

Before using this method the method SamplingComponent.initForSampling(int) should be called.

Specified by:

drawParametersFromStatistic in interface SamplingFromStatistic

Throws:

Exception - if there is a problem with drawing the parameters, the initialization, ...

See Also:

SamplingComponent.initForSampling(int), SamplingComponent.acceptParameters()

getLogGammaScoreFromStatistic

public double getLogGammaScoreFromStatistic()

Description copied from interface: SamplingEmission

This method calculates a score for the current statistics, which is independent from the current parameters In general the gamma-score is a product of gamma-functions parameterized with the current statistics

Specified by:

getLogGammaScoreFromStatistic in interface SamplingEmission

Returns:

the logarithm of the gamma-score for the current statistics

acceptParameters

public void acceptParameters()
                      throws IOException

Description copied from interface: SamplingComponent

This methods accepts the drawn parameters. Internally the drawn parameters should be saved (to a file).

Specified by:

acceptParameters in interface SamplingComponent

Throws:

IOException - if the file could not be handled correctly

getLogPosteriorFromStatistic

public double getLogPosteriorFromStatistic()

Description copied from interface: SamplingFromStatistic

This method calculates the a-posteriori probability for the current statistics

Specified by:

getLogPosteriorFromStatistic in interface SamplingFromStatistic

Returns:

the logarithm of the a-posteriori probability

extendSampling

public void extendSampling(int start,
                           boolean append)
                    throws IOException

Description copied from interface: SamplingComponent

This method allows to extend a sampling.

Specified by:

extendSampling in interface SamplingComponent

Parameters:

start - the index of the sampling

append - whether to append the sampled parameters to an existing file or to overwrite the file

Throws:

IOException - if the file could not be handled correctly

initForSampling

public void initForSampling(int starts)
                     throws IOException

Description copied from interface: SamplingComponent

This method initializes the instance for the sampling. For instance this method can be used to create new files where all parameter sets are stored.

Specified by:

initForSampling in interface SamplingComponent

Parameters:

starts - the number of different sampling starts that will be done

Throws:

IOException - if something went wrong

See Also:

File.createTempFile(String, String, java.io.File )

isInSamplingMode

public boolean isInSamplingMode()

Description copied from interface: SamplingComponent

This method returns true if the object is currently used in a sampling, otherwise false.

Specified by:

isInSamplingMode in interface SamplingComponent

Returns:

true if the object is currently used in a sampling, otherwise false

parseNextParameterSet

public boolean parseNextParameterSet()

Description copied from interface: SamplingComponent

This method allows the user to parse the next set of parameters (from a file).

Specified by:

parseNextParameterSet in interface SamplingComponent

Returns:

true if the parameters could be parsed, otherwise false

See Also:

SamplingComponent.parseParameterSet(int, int)

parseParameterSet

public boolean parseParameterSet(int start,
                                 int n)
                          throws IOException

Description copied from interface: SamplingComponent

This method allows the user to parse the set of parameters with index n of a certain sampling (from a file). The internal numbering should start with 0. The parameter set with index 0 is the initial (random) parameter set. It is recommended that a series of parameter sets is accessed by the following lines:

for( sampling = 0; sampling < numSampling; sampling++ )
{

boolean b = parseParameterSet( sampling, n );
while( b )
{
//do something
b = parseNextParameterSet();
} }


Specified by:

parseParameterSet in interface SamplingComponent

Parameters:

start - the index of the sampling

n - the index of the parameter set

Returns:

true if the parameter set could be parsed

Throws:

IOException

See Also:

SamplingComponent.parseNextParameterSet()

samplingStopped

public void samplingStopped()
                     throws IOException

Description copied from interface: SamplingComponent

This method is the opposite of the method SamplingComponent.extendSampling(int, boolean). It can be used for closing any streams of writer, ...

Specified by:

samplingStopped in interface SamplingComponent

Throws:

IOException - if something went wrong

See Also:

SamplingComponent.extendSampling(int, boolean)

finalize

protected void finalize()
                 throws Throwable

Overrides:

finalize in class Object

Throws:

Throwable

getNodeShape

public String getNodeShape(boolean forward)

Description copied from interface: Emission

Returns the graphviz string for the shape of the node.

Specified by:

getNodeShape in interface Emission

Parameters:

forward - if this emission is used on the forward strand

Returns:

the shape

getNodeLabel

public String getNodeLabel(double weight,
                           String name,
                           NumberFormat nf)

Description copied from interface: Emission

Returns the graphviz label of the node containing this emission.

Specified by:

getNodeLabel in interface Emission

Parameters:

weight - the weight of the node which is represented by the color of the node, or -1 for no representation, i.e., white background

name - the name of the state using this emission

nf - the NumberFormat for formatting the textual representation of this emission

Returns:

the label

setLinear

public void setLinear(boolean linear)

If set to true, the probabilities are mapped to colors by directly, otherwise a logistic mapping is used to emphasize deviations from the uniform distribution.

Parameters:

linear - map probabilities linear

fillSamplingGroups

public void fillSamplingGroups(int parameterOffset,
                               LinkedList<int[]> list)

Description copied from interface: DifferentiableEmission

Adds the groups of indexes of those parameters of this emission that should be sampled together in one step of a grouped sampling procedure, each as an int[], into list. In most cases, one group should contain the parameters that are living on a common simplex. The internal indexes of the parameters are incremeneted by an external parameterOffset

Specified by:

fillSamplingGroups in interface DifferentiableEmission

Parameters:

parameterOffset - the external parameter offset

list - the list of sampling groups

getNumberOfParameters

public int getNumberOfParameters()

Description copied from interface: DifferentiableEmission

Returns the number of parameters of this emission.

Specified by:

getNumberOfParameters in interface DifferentiableEmission

Returns:

the number of parameters

getSizeOfEventSpace

public int getSizeOfEventSpace()

Description copied from interface: DifferentiableEmission

Returns the size of the event space, i.e., the number of possible outcomes, for the random variables of this emission

Specified by:

getSizeOfEventSpace in interface DifferentiableEmission

Returns:

the size of the event space