com.emphysic.myriad.core.data.roi

Class AdaptiveSGDROIFinder

java.lang.Object

com.emphysic.myriad.core.data.roi.MLROIConfFinder

com.emphysic.myriad.core.data.roi.AdaptiveSGDROIFinder

All Implemented Interfaces:

MLROIFinder, ROIFinder, ROIProbability, ObjectMap, com.esotericsoftware.kryo.KryoSerializable

public class AdaptiveSGDROIFinder
extends MLROIConfFinder

AdaptiveSGDROIFinder - machine learning ROI detector that locates regions of interest by maintaining a pool of SGD classifiers.

Field Summary

Fields inherited from class com.emphysic.myriad.core.data.roi.MLROIConfFinder

confThr

Constructor Summary

Constructors

Constructor and Description
AdaptiveSGDROIFinder() Create a flaw finder with 2 categories - is / isn't a flaw, and a linear combination of Laplacian and Gaussian regularization.
AdaptiveSGDROIFinder(int numCats, org.apache.mahout.classifier.sgd.PriorFunction priorFunction) Creates a new AdaptiveSGDROIFinder with the specified number of categories and the specified regularization function.
AdaptiveSGDROIFinder(int numCats, org.apache.mahout.classifier.sgd.PriorFunction priorFunction, int threadCount, int poolSize)

Method Summary

Modifier and Type	Method and Description
org.apache.mahout.math.Vector	classify(org.apache.mahout.math.DenseVector d)
void	close() Shuts down the worker pool used during training and best model determination.
org.apache.mahout.classifier.sgd.AdaptiveLogisticRegression	getModel() Returns the current meta-model.
int	getNumCategories() Return the number of categories for this model.
int	getNumFeatures() Return the number of features for this model.
java.util.Map<java.lang.String,java.lang.Object>	getObjectMap() Creates a map of the important fields for the instance, suitable for serialization.
int	getPoolSize() Returns the number of SGD learners in the meta-model.
org.apache.mahout.classifier.sgd.PriorFunction	getPriorFunction() Returns the regularization function used to limit overfitting.
long	getSerializationVersion() Returns the current version of the serialization format.
int	getThreadCount() Returns the number of threads used in training the meta-model learners
int	getVersion() Returns the current class version.
void	initCurrentVersion(java.util.Map<java.lang.String,java.lang.Object> objectMap) Initializes an instance with a current-version object graph.
boolean	isROI(Dataset dataset) Predict if a sample contains a region of interest.
boolean	isROI(double[] data) Predict if a sample contains a region of interest.
void	legacyRead(com.esotericsoftware.kryo.Kryo kryo, com.esotericsoftware.kryo.io.Input input)
void	legacyWrite(com.esotericsoftware.kryo.Kryo kryo, com.esotericsoftware.kryo.io.Output output)
double	negativeClass() Returns the numeric value of the negative class of a two-category model.
double	positiveClass() Returns the numeric value of the positive class of a two-category model.
void	setModel(org.apache.mahout.classifier.sgd.AdaptiveLogisticRegression model) Sets the current meta-model
void	setNumCategories(int numCategories) Sets the number of categories for this model.
void	setNumFeatures(int numFeatures) Sets the number of features for this model.
void	setPoolSize(int poolSize) Sets the number of SGD learners in the meta-model.
void	setPriorFunction(org.apache.mahout.classifier.sgd.PriorFunction priorFunction) Sets the regularization function used to limit overfitting.
void	setThreadCount(int threadCount) Sets the number of threads to use during training.
void	train(double[][] X, int[] y) Trains the current model with additional samples, creating the model if required.

Methods inherited from class com.emphysic.myriad.core.data.roi.MLROIConfFinder

getConfidenceThreshold, predict, predict, setConfidenceThreshold, writableWriteToBytes

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Methods inherited from interface com.emphysic.myriad.core.data.roi.ROIFinder

fromFile

Methods inherited from interface com.emphysic.myriad.core.data.util.ObjectMap

init, initPreviousVersion, initUnknownVersion, load, read, save, write

Methods inherited from interface com.emphysic.myriad.core.data.roi.ROIProbability

predict_proba, predict_proba

Constructor Detail

AdaptiveSGDROIFinder

public AdaptiveSGDROIFinder(int numCats,
                            org.apache.mahout.classifier.sgd.PriorFunction priorFunction)

Creates a new AdaptiveSGDROIFinder with the specified number of categories and the specified regularization function. Uses all available CPU cores and 20 SGD learners.

Parameters:

numCats - number of categories

priorFunction - prior function to use

AdaptiveSGDROIFinder

public AdaptiveSGDROIFinder(int numCats,
                            org.apache.mahout.classifier.sgd.PriorFunction priorFunction,
                            int threadCount,
                            int poolSize)

Parameters:

numCats - number of labels in the data (e.g. 2 for is/is not a flaw)

priorFunction - Regularization function to penalize complex models

threadCount - number of cores to use during training. Defaults to number of cores on system.

poolSize - number of learners to train. Defaults to 20.

AdaptiveSGDROIFinder

public AdaptiveSGDROIFinder()

Create a flaw finder with 2 categories - is / isn't a flaw, and a linear combination of Laplacian and Gaussian regularization. Twenty learners are trained, using all the available cores on the system.

Method Detail

train

public void train(double[][] X,
                  int[] y)
           throws java.lang.Exception

Trains the current model with additional samples, creating the model if required.

Parameters:

X - N examples with M features per example

y - N labels for the N examples in X

Throws:

java.lang.Exception - if an error occurs

classify

public org.apache.mahout.math.Vector classify(org.apache.mahout.math.DenseVector d)

isROI

public boolean isROI(double[] data)

Predict if a sample contains a region of interest.

Parameters:

data - raw data to examine

Returns:

true if the sample appears to contain an ROI, false otherwise.

isROI

public boolean isROI(Dataset dataset)

Predict if a sample contains a region of interest.

Parameters:

dataset - data to examine

Returns:

true if the sample appears to contain an ROI, false otherwise.

getSerializationVersion

public long getSerializationVersion()

Description copied from interface: ObjectMap

Returns the current version of the serialization format. Increments when the serialization format changes, i.e. backwards compatibility is broken.

Returns:

serialization version

getVersion

public int getVersion()

Description copied from interface: ObjectMap

Returns the current class version. Class version increments when the class changes but doesn't necessarily break compatibility with serialization.

Returns:

class version

getObjectMap

public java.util.Map<java.lang.String,java.lang.Object> getObjectMap()

Description copied from interface: ObjectMap

Creates a map of the important fields for the instance, suitable for serialization.

Returns:

"object map" of the instance

initCurrentVersion

public void initCurrentVersion(java.util.Map<java.lang.String,java.lang.Object> objectMap)

Description copied from interface: ObjectMap

Initializes an instance with a current-version object graph.

Parameters:

objectMap - object graph for initialization

getNumCategories

public int getNumCategories()

Return the number of categories for this model. Typically 2 for does / does not contain a region of interest.

Returns:

number of categories

setNumCategories

public void setNumCategories(int numCategories)

Sets the number of categories for this model. Note that it may be necessary to retrain the model if this number is changed.

Parameters:

numCategories - number of categories

getNumFeatures

public int getNumFeatures()

Return the number of features for this model. In Myriad, this is equivalent to the size of the data the model was trained on i.e. #Dataset.getSize().

Returns:

number of features

setNumFeatures

public void setNumFeatures(int numFeatures)

Sets the number of features for this model. Note that changes to this number will invalidate the current model.

Parameters:

numFeatures - number of features

getPriorFunction

public org.apache.mahout.classifier.sgd.PriorFunction getPriorFunction()

Returns the regularization function used to limit overfitting.

Returns:

regularization function.

setPriorFunction

public void setPriorFunction(org.apache.mahout.classifier.sgd.PriorFunction priorFunction)

Sets the regularization function used to limit overfitting.

Parameters:

priorFunction - new regularization function.

getModel

public org.apache.mahout.classifier.sgd.AdaptiveLogisticRegression getModel()

Returns the current meta-model.

Returns:

meta-model

setModel

public void setModel(org.apache.mahout.classifier.sgd.AdaptiveLogisticRegression model)

Sets the current meta-model

Parameters:

model - new meta-model

positiveClass

public double positiveClass()

Returns the numeric value of the positive class of a two-category model.

Returns:

positive class label

negativeClass

public double negativeClass()

Returns the numeric value of the negative class of a two-category model.

Returns:

negative class label

close

public void close()

Shuts down the worker pool used during training and best model determination.

getThreadCount

public int getThreadCount()

Returns the number of threads used in training the meta-model learners

Returns:

number of threads (cores)

setThreadCount

public void setThreadCount(int threadCount)

Sets the number of threads to use during training.

Parameters:

threadCount - new number of threads (cores)

getPoolSize

public int getPoolSize()

Returns the number of SGD learners in the meta-model.

Returns:

number of SGD learners

setPoolSize

public void setPoolSize(int poolSize)

Sets the number of SGD learners in the meta-model.

Parameters:

poolSize - new number of SGD learners

legacyWrite

public void legacyWrite(com.esotericsoftware.kryo.Kryo kryo,
                        com.esotericsoftware.kryo.io.Output output)

legacyRead

public void legacyRead(com.esotericsoftware.kryo.Kryo kryo,
                       com.esotericsoftware.kryo.io.Input input)