com.pervasive.datarush.analytics.naivebayes.predictor

Class NaiveBayesPredictor

java.lang.Object

com.pervasive.datarush.operators.AbstractLogicalOperator

com.pervasive.datarush.operators.StreamingOperator

com.pervasive.datarush.operators.ExecutableOperator

com.pervasive.datarush.analytics.util.AbstractPredictor

com.pervasive.datarush.analytics.naivebayes.predictor.NaiveBayesPredictor

All Implemented Interfaces:

LogicalOperator

public final class NaiveBayesPredictor
extends AbstractPredictor

Operator responsible for predicting outcomes based on a Naive Bayes PMML model. The base algorithm used is specified here here, with the following differences:

1. Provides the ability to predict based on numerical data. For numerical data, we compute probability based on the assumption of a Gaussian distribution.
2. We use Laplace smoothing in place of the "threshold" parameter.
3. We provide an option to count missing values. If selected, missing values are treated like any other single distinct value. Probability is calculated in terms of the ration of missing to non-missing.
4. Calculation is performed in terms of log-likelihood rather than likelihood.

Constructor Summary

Constructors

Constructor and Description
NaiveBayesPredictor() Predicts an output based on a model and a set of training data.

Method Summary

Methods

Modifier and Type	Method and Description
protected void	computeMetadata(StreamingMetadataContext ctx) Default implementation of computeMetadata.
protected void	execute(PMMLModel pmml, RecordValued input, ScalarSettable[] predictedFields) Called to perform prediction.
double	getLaplaceCorrector() Returns the Laplace corrector to be used.
RecordPort	getOutput() Returns a record port consisting of the input plus predicted values appended.
String	getProbabilityPrefix() Gets the field name prefix to use for probabilities.
String	getWinnerField() Gets the name of the winner field to output.
boolean	isAppendProbabilities() Returns whether to include probabilities in the prediction.
boolean	isIgnoreMissingValues() Returns whether to ignore missing values.
protected RecordTokenType	predictedType(PMMLModelSpec spec) Given the model spec, returns the predicted type.
void	setAppendProbabilities(boolean appendProbabilities) Sets whether to include probabilities in the prediction.
void	setIgnoreMissingValues(boolean ignoreMissingValues) Sets whether to ignore missing values.
void	setLaplaceCorrector(double laplaceCorrector) Sets the Laplace corrector to be used.
void	setProbabilityPrefix(String probabilityPrefix) Sets the field name prefix to use for probabilities.
void	setWinnerField(String winnerField) Sets the name of the winner field to output.

Methods inherited from class com.pervasive.datarush.analytics.util.AbstractPredictor

execute, getInput, getModel, pushPrediction, stepNext

Methods inherited from class com.pervasive.datarush.operators.ExecutableOperator

cloneForExecution, getNumInputCopies, getPortSettings, handleInactiveOutput

Methods inherited from class com.pervasive.datarush.operators.AbstractLogicalOperator

disableParallelism, getInputPorts, getOutputPorts, newInput, newInput, newOutput, newRecordInput, newRecordInput, newRecordOutput, notifyError

Methods inherited from class java.lang.Object

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

Constructor Detail

NaiveBayesPredictor

public NaiveBayesPredictor()

Predicts an output based on a model and a set of training data. Following construction via the default constructor, no additional configuration is required.

Method Detail

getOutput

public RecordPort getOutput()

Returns a record port consisting of the input plus predicted values appended. The record has the following fields:

1. winner: The target value of highest probability. The name "winner" is the default; this is configurable via the property winnerField.
2. probability_targetValue (optional): The probability of the named targetValue. The prefix "probability_" is the default; this is configurable via the property probabilityPrefix.

Overrides:

getOutput in class AbstractPredictor

Returns:

a record flow of predicted values and their probabilities.

getWinnerField

public String getWinnerField()

Gets the name of the winner field to output. This is "winner" by-default.

Returns:

the name of the winner field to output.

setWinnerField

public void setWinnerField(String winnerField)

Sets the name of the winner field to output. This is "winner" by-default.

Parameters:

winnerField - the name of the winner field to output.

getProbabilityPrefix

public String getProbabilityPrefix()

Gets the field name prefix to use for probabilities. This is "probability_" by-default.

Returns:

the field name prefix to use for probabilities.

setProbabilityPrefix

public void setProbabilityPrefix(String probabilityPrefix)

Sets the field name prefix to use for probabilities. This is "probability_" by-default.

Parameters:

probabilityPrefix - the field name prefix to use for probabilities.

setIgnoreMissingValues

public void setIgnoreMissingValues(boolean ignoreMissingValues)

Sets whether to ignore missing values. If set to true, missing values are ignored for the purposes of prediction; otherwise missing values are considered when calculating probability distribution. Defaults to true.

Parameters:

ignoreMissingValues - whether to ignore missing values

isIgnoreMissingValues

public boolean isIgnoreMissingValues()

Returns whether to ignore missing values. If set to true, missing values are ignored for the purposes of prediction; otherwise missing values are considered when calculating probability distribution. Defaults to true.

Returns:

whether to ignore missing values

getLaplaceCorrector

public final double getLaplaceCorrector()

Returns the Laplace corrector to be used. The Laplace corrector is a way to handle "zero" counts in the training data. Otherwise a value that was never observed in the training data results in zero probability. The default of 0.0 means no correction.

NOTE: The "threshold" value specified in the PMML model will always be ignored in favor of the Laplace corrector specified on NaiveBayesPredictor.

Returns:

the Laplace corrector to be used.

setLaplaceCorrector

public final void setLaplaceCorrector(double laplaceCorrector)

Sets the Laplace corrector to be used. The Laplace corrector is a way to handle "zero" counts in the training data. Otherwise a value that was never observed in the training data results in zero probability. The default of 0.0 means no correction.

NOTE: The "threshold" value specified in the PMML model will always be ignored in favor of the Laplace corrector specified on NaiveBayesPredictor.

Parameters:

laplaceCorrector - the Laplace corrector to be used.

isAppendProbabilities

public boolean isAppendProbabilities()

Returns whether to include probabilities in the prediction. This is true by default.

Returns:

whether to include probabilities in the prediction

setAppendProbabilities

public void setAppendProbabilities(boolean appendProbabilities)

Sets whether to include probabilities in the prediction. This is true by default.

Parameters:

appendProbabilities - whether to include probabilities in the prediction

computeMetadata

protected void computeMetadata(StreamingMetadataContext ctx)

Description copied from class: AbstractPredictor

Default implementation of computeMetadata.

1. Output type is set to input type plus predictedType
2. Input data ordering ( if ordered ) is preserved
3. Input data partitioning ( if partitioned ) is preserved

Overrides:

computeMetadata in class AbstractPredictor

Parameters:

ctx - the context

predictedType

protected RecordTokenType predictedType(PMMLModelSpec spec)

Description copied from class: AbstractPredictor

Given the model spec, returns the predicted type. This should not include the input type ( the input is automatically prepended to the type that is returned )

Specified by:

predictedType in class AbstractPredictor

Parameters:

spec - the model metadata

Returns:

the predicted type

execute

protected void execute(PMMLModel pmml,
           RecordValued input,
           ScalarSettable[] predictedFields)

Description copied from class: AbstractPredictor

Called to perform prediction. Subclasses are expected to loop over the input by calling AbstractPredictor.stepNext(). For each row of input, subclasses should first set the predicted values in the predictedFields array and then invoke AbstractPredictor.pushPrediction(). Subclasses should not invoke pushEndOfData since that is automatically handled by the base class.

Specified by:

execute in class AbstractPredictor

Parameters:

pmml - The input PMML model

input - The input data

predictedFields - An array of fields that reference the predicted field locations. The array positionally corresponds to the type returned by AbstractPredictor.predictedType(PMMLModelSpec).