org.pentaho.di.forecasting

Class WekaForecastingData

java.lang.Object

org.pentaho.di.trans.step.BaseStepData

org.pentaho.di.forecasting.WekaForecastingData

All Implemented Interfaces:

org.pentaho.di.trans.step.StepDataInterface

public class WekaForecastingData
extends org.pentaho.di.trans.step.BaseStepData
implements org.pentaho.di.trans.step.StepDataInterface

Holds temporary data and has routines for loading serialized models.

Version:

$Revision$

Author:

Mark Hall (mhall{[at]}pentaho{[dot]}com)

Nested Class Summary

Nested classes/interfaces inherited from class org.pentaho.di.trans.step.BaseStepData

org.pentaho.di.trans.step.BaseStepData.StepExecutionStatus

Field Summary

Fields

Modifier and Type	Field and Description
protected org.pentaho.di.core.row.RowMetaInterface	m_outputRowMeta
static int	NO_MATCH
static int	TYPE_MISMATCH

Constructor Summary

Constructors

Constructor and Description
WekaForecastingData()

Method Summary

Modifier and Type	Method and Description
protected weka.core.Instance	constructInstance(org.pentaho.di.core.row.RowMetaInterface inputMeta, Object[] inputRow, int[] mappingIndexes, WekaForecastingModel model) Helper method that constructs an Instance to input to the Weka model based on incoming Kettle fields and pre-constructed attribute-to-field mapping data.
static int[]	findMappings(weka.core.Instances header, org.pentaho.di.core.row.RowMetaInterface inputRowMeta) Finds a mapping between the attributes that a forecasting model has been trained with and the incoming Kettle row format.
void	fixTypesForTargets(Object[] row, List<String> targetFields, org.pentaho.di.core.row.RowMetaInterface inputRowMeta) Fixes the type of forecasted fields (if necessary).
List<Object[]>	generateForecast(org.pentaho.di.core.row.RowMetaInterface inputMeta, org.pentaho.di.core.row.RowMetaInterface outputMeta, WekaForecastingMeta meta, List<Object[]> overlayData, org.pentaho.di.trans.TransMeta transMeta, PrintStream... progress) Generates a forecast given a forecasting model (sourced from the meta object).
org.pentaho.di.core.row.RowMetaInterface	getOutputRowMeta() Get the meta data for the output format
static WekaForecastingModel	loadSerializedModel(File modelFile, org.pentaho.di.core.logging.LogChannelInterface log) Loads a serialized model.
static void	saveSerializedModel(WekaForecastingModel wsm, File saveTo)
void	setOutputRowMeta(org.pentaho.di.core.row.RowMetaInterface rmi) Set the meta data for the output format
boolean	sortCheck(weka.classifiers.timeseries.TSForecaster forecaster, weka.core.Instances data)

Methods inherited from class org.pentaho.di.trans.step.BaseStepData

getStatus, isDisposed, isEmpty, isFinished, isIdle, isInitialising, isRunning, isStopped, setStatus

Methods inherited from class java.lang.Object

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

Methods inherited from interface org.pentaho.di.trans.step.StepDataInterface

getStatus, isDisposed, isEmpty, isFinished, isIdle, isInitialising, isRunning, setStatus

Field Detail

NO_MATCH

public static final int NO_MATCH

See Also:

Constant Field Values

TYPE_MISMATCH

public static final int TYPE_MISMATCH

See Also:

Constant Field Values

m_outputRowMeta

protected org.pentaho.di.core.row.RowMetaInterface m_outputRowMeta

Constructor Detail

WekaForecastingData

public WekaForecastingData()

Method Detail

getOutputRowMeta

public org.pentaho.di.core.row.RowMetaInterface getOutputRowMeta()

Get the meta data for the output format

Returns:

a RowMetaInterface value

setOutputRowMeta

public void setOutputRowMeta(org.pentaho.di.core.row.RowMetaInterface rmi)

Set the meta data for the output format

Parameters:

rmi - a RowMetaInterface value

fixTypesForTargets

public void fixTypesForTargets(Object[] row,
                               List<String> targetFields,
                               org.pentaho.di.core.row.RowMetaInterface inputRowMeta)
                        throws org.pentaho.di.core.exception.KettleException

Fixes the type of forecasted fields (if necessary). A forecaster forecasts target values as doubles. If incoming target fields values are non-double numeric types (as they might be for historical priming rows), then values need to be converted to Double to match the output row meta data.

Parameters:

row - row to check

targetFields - list of target fields predicted by the forecaster

inputRowMeta - the input row meta data

Throws:

org.pentaho.di.core.exception.KettleException - if a problem occurs.

sortCheck

public boolean sortCheck(weka.classifiers.timeseries.TSForecaster forecaster,
                         weka.core.Instances data)
                  throws org.pentaho.di.core.exception.KettleException

Throws:

org.pentaho.di.core.exception.KettleException

loadSerializedModel

public static WekaForecastingModel loadSerializedModel(File modelFile,
                                                       org.pentaho.di.core.logging.LogChannelInterface log)
                                                throws Exception

Loads a serialized model. Models can either be binary serialized Java objects, objects deep-serialized to xml, or PMML.

Parameters:

modelFile - a File value

Returns:

the model

Throws:

Exception - if there is a problem laoding the model.

saveSerializedModel

public static void saveSerializedModel(WekaForecastingModel wsm,
                                       File saveTo)
                                throws Exception

Throws:

Exception

findMappings

public static int[] findMappings(weka.core.Instances header,
                                 org.pentaho.di.core.row.RowMetaInterface inputRowMeta)

Finds a mapping between the attributes that a forecasting model has been trained with and the incoming Kettle row format. Returns an array of indices, where the element at index 0 of the array is the index of the Kettle field that corresponds to the first attribute in the Instances structure, the element at index 1 is the index of the Kettle fields that corresponds to the second attribute, ...

Parameters:

header - the Instances header

inputRowMeta - the meta data for the incoming rows

Returns:

the mapping as an array of integer indices

generateForecast

public List<Object[]> generateForecast(org.pentaho.di.core.row.RowMetaInterface inputMeta,
                                       org.pentaho.di.core.row.RowMetaInterface outputMeta,
                                       WekaForecastingMeta meta,
                                       List<Object[]> overlayData,
                                       org.pentaho.di.trans.TransMeta transMeta,
                                       PrintStream... progress)
                                throws Exception

Generates a forecast given a forecasting model (sourced from the meta object).

Parameters:

inputMeta - the incoming row meta data

outputMeta - the outgoing row meta data

meta - the forecasting meta

overlayData - a list of rows for future time steps (in the same format as the incoming rows) containing values for "overlay" fields. May be null if overlay data is not in use.

Returns:

a List of rows containing the forecast.

Throws:

Exception - if a problem occurs.

constructInstance

protected weka.core.Instance constructInstance(org.pentaho.di.core.row.RowMetaInterface inputMeta,
                                               Object[] inputRow,
                                               int[] mappingIndexes,
                                               WekaForecastingModel model)

Helper method that constructs an Instance to input to the Weka model based on incoming Kettle fields and pre-constructed attribute-to-field mapping data.

Parameters:

inputMeta - a RowMetaInterface value

inputRow - an Object value

mappingIndexes - an int value

model - a WekaScoringModel value

Returns:

an Instance value