weka.classifiers.timeseries

Class WekaForecaster

java.lang.Object

weka.classifiers.timeseries.AbstractForecaster

weka.classifiers.timeseries.WekaForecaster

All Implemented Interfaces:

Serializable, ConfidenceIntervalForecaster, IncrementallyPrimeable, OverlayForecaster, TSLagUser, TSForecaster, weka.core.CommandlineRunnable, weka.core.OptionHandler

public class WekaForecaster
extends AbstractForecaster
implements TSLagUser, ConfidenceIntervalForecaster, OverlayForecaster, IncrementallyPrimeable, weka.core.OptionHandler, Serializable

Class that implements time series forecasting using a Weka regression scheme. Makes use of the TSLagMaker class to handle all lagged attribute creation, periodic attributes etc.

Version:

$Revision: 52593 $

Author:

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

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
WekaForecaster()

Method Summary

Methods

Modifier and Type	Method and Description
void	addCustomPeriodic(String customPeriodic) Add a custom date-derived periodic attribute
boolean	baseModelHasSerializer() Check whether the base learner requires special serialization
void	buildForecaster(weka.core.Instances insts, PrintStream... progress) Builds a new forecasting model using the supplied training data.
void	clearCustomPeriodics() clear the list of custom date-derived periodic attributes
void	clearPreviousState() Reset model state.
List<List<weka.classifiers.evaluation.NumericPrediction>>	forecast(int numSteps, weka.core.Instances overlay, PrintStream... progress) Produce a forecast for the target field(s).
List<List<weka.classifiers.evaluation.NumericPrediction>>	forecast(int numSteps, PrintStream... progress) Produce a forecast for the target field(s).
String	getAlgorithmName() Provides a short name that describes the underlying algorithm in some way.
weka.classifiers.Classifier	getBaseForecaster() Get the base Weka regression scheme being used to make forecasts
int	getCalculateConfIntervalsForForecasts() Return the number of steps for which confidence intervals will be computed.
double	getConfidenceLevel() Get the confidence level in use for computing confidence intervals.
String[]	getOptions() Gets the current settings of this Forecaster.
String	getOverlayFields() Get a comma-separated list of fields that considered to be overlay fields
List<Object>	getPreviousState() Get the last set state of the model.
TSLagMaker	getTSLagMaker() Get the TSLagMaker that we are using.
boolean	isProducingConfidenceIntervals() Returns true if this forecaster is computing confidence limits for some or all of its future forecasts (i.e.
boolean	isUsingOverlayData() Returns true if overlay data has been used to train this forecaster, and thus is expected to be supplied for future time steps when making a forecast.
Enumeration<weka.core.Option>	listOptions() Returns an enumeration describing the available options.
void	loadBaseModel(String filepath) Load serialized classifier
void	loadSerializedState(String filepath) Load serialized model state
static void	main(String[] args) Main method for running this class from the command line
void	primeForecaster(weka.core.Instances insts) Supply the (potentially) trained model with enough historical data, up to and including the current time point, in order to produce a forecast.
void	primeForecasterIncremental(weka.core.Instance inst) Update the priming information incrementally, i.e.
void	reset() Reset the forecaster.
void	saveBaseModel(String filepath) Save underlying classifier
void	serializeState(String filepath) Serialize model state
void	setBaseForecaster(weka.classifiers.Classifier f) Set the base Weka regression scheme to use.
void	setCalculateConfIntervalsForForecasts(int steps) Set the number of steps for which to compute confidence intervals for.
void	setConfidenceLevel(double confLevel) Set the confidence level for confidence intervals.
void	setFieldsToForecast(String fieldsToForecast) Set the names of the fields/attributes in the data to forecast.
void	setOptions(String[] options) Set the options for the forecaster
void	setOverlayFields(String overlayFields) Set the fields to consider as overlay fields
void	setPreviousState(List<Object> previousState) Load state into model.
void	setTSLagMaker(TSLagMaker lagMaker) Set the TSLagMaker to use.
String	toString()
boolean	usesState() Check whether the base learner requires operations regarding state

Methods inherited from class weka.classifiers.timeseries.AbstractForecaster

getFieldsToForecast, makeCopy, postExecution, preExecution, run, runForecaster, stringToList

Methods inherited from class java.lang.Object

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

Constructor Detail

WekaForecaster

public WekaForecaster()

Method Detail

main

public static void main(String[] args)

Main method for running this class from the command line

Parameters:

args - general and scheme-specific command line arguments

baseModelHasSerializer

public boolean baseModelHasSerializer()

Check whether the base learner requires special serialization

Specified by:

baseModelHasSerializer in interface TSForecaster

Returns:

true if base learner requires special serialization, false otherwise

saveBaseModel

public void saveBaseModel(String filepath)
                   throws Exception

Save underlying classifier

Specified by:

saveBaseModel in interface TSForecaster

Parameters:

filepath - the path of the file to save the base model to

Throws:

Exception

loadBaseModel

public void loadBaseModel(String filepath)
                   throws Exception

Load serialized classifier

Specified by:

loadBaseModel in interface TSForecaster

Parameters:

filepath - the path of the file to load the base model from

Throws:

Exception

serializeState

public void serializeState(String filepath)
                    throws Exception

Serialize model state

Specified by:

serializeState in interface TSForecaster

Parameters:

filepath - the path of the file to save the model state to

Throws:

Exception

loadSerializedState

public void loadSerializedState(String filepath)
                         throws Exception

Load serialized model state

Specified by:

loadSerializedState in interface TSForecaster

Parameters:

filepath - the path of the file to save the model state from

Throws:

Exception

usesState

public boolean usesState()

Check whether the base learner requires operations regarding state

Specified by:

usesState in interface TSForecaster

Returns:

true if base learner uses state-based predictions, false otherwise

clearPreviousState

public void clearPreviousState()

Reset model state.

Specified by:

clearPreviousState in interface TSForecaster

setPreviousState

public void setPreviousState(List<Object> previousState)

Load state into model.

Specified by:

setPreviousState in interface TSForecaster

getPreviousState

public List<Object> getPreviousState()

Get the last set state of the model.

Specified by:

getPreviousState in interface TSForecaster

Returns:

the state of the model to be used in next prediction

getAlgorithmName

public String getAlgorithmName()

Provides a short name that describes the underlying algorithm in some way.

Specified by:

getAlgorithmName in interface TSForecaster

Returns:

a short description of this forecaster.

getTSLagMaker

public TSLagMaker getTSLagMaker()

Get the TSLagMaker that we are using. All options pertaining to lag creation, periodic attributes etc. are set via the lag maker.

Specified by:

getTSLagMaker in interface TSLagUser

Returns:

the TSLagMaker that we are using.

setTSLagMaker

public void setTSLagMaker(TSLagMaker lagMaker)

Set the TSLagMaker to use. All options pertaining to lag creation, periodic attributes etc. are set via the lag maker.

Specified by:

setTSLagMaker in interface TSLagUser

Parameters:

lagMaker - the TSLagMaker to use.

listOptions

public Enumeration<weka.core.Option> listOptions()

Returns an enumeration describing the available options.

Specified by:

listOptions in interface weka.core.OptionHandler

Returns:

an enumeration of all the available options.

getOptions

public String[] getOptions()

Gets the current settings of this Forecaster.

Specified by:

getOptions in interface weka.core.OptionHandler

Returns:

an array of strings suitable for passing to setOptions

setOptions

public void setOptions(String[] options)
                throws Exception

Set the options for the forecaster

Specified by:

setOptions in interface weka.core.OptionHandler

Parameters:

options - an array of options

Throws:

Exception - if unknown options are supplied

addCustomPeriodic

public void addCustomPeriodic(String customPeriodic)

Add a custom date-derived periodic attribute

Parameters:

customPeriodic - the string definition of the custom date derived periodic attribute to add

clearCustomPeriodics

public void clearCustomPeriodics()

clear the list of custom date-derived periodic attributes

setFieldsToForecast

public void setFieldsToForecast(String fieldsToForecast)
                         throws Exception

Set the names of the fields/attributes in the data to forecast.

Specified by:

setFieldsToForecast in interface TSForecaster

Overrides:

setFieldsToForecast in class AbstractForecaster

Parameters:

fieldsToForecast - a list of names of fields to forecast

Throws:

Exception - if a field(s) can't be found, or if multiple fields are specified and this forecaster can't predict multiple fields.

getOverlayFields

public String getOverlayFields()

Get a comma-separated list of fields that considered to be overlay fields

Specified by:

getOverlayFields in interface OverlayForecaster

Returns:

a list of field names

setOverlayFields

public void setOverlayFields(String overlayFields)
                      throws Exception

Set the fields to consider as overlay fields

Specified by:

setOverlayFields in interface OverlayForecaster

Parameters:

overlayFields - a comma-separated list of fieldnames

Throws:

Exception - if there is a problem setting the overlay fields

getCalculateConfIntervalsForForecasts

public int getCalculateConfIntervalsForForecasts()

Return the number of steps for which confidence intervals will be computed.

Specified by:

getCalculateConfIntervalsForForecasts in interface ConfidenceIntervalForecaster

Returns:

the number of steps for which confidence intervals will be computed.

setCalculateConfIntervalsForForecasts

public void setCalculateConfIntervalsForForecasts(int steps)

Set the number of steps for which to compute confidence intervals for. E.g. a value of 5 means that confidence bounds will be computed for 1-step-ahead predictions, 2-step-ahead predictions, ..., 5-step-ahead predictions.

Specified by:

setCalculateConfIntervalsForForecasts in interface ConfidenceIntervalForecaster

Parameters:

steps - the number of steps for which to compute confidence intervals for.

isProducingConfidenceIntervals

public boolean isProducingConfidenceIntervals()

Returns true if this forecaster is computing confidence limits for some or all of its future forecasts (i.e. getCalculateConfIntervalsForForecasts() > 0).

Specified by:

isProducingConfidenceIntervals in interface ConfidenceIntervalForecaster

Returns:

true if confidence limits will be produced for some or all of its future forecasts.

getConfidenceLevel

public double getConfidenceLevel()

Get the confidence level in use for computing confidence intervals.

Specified by:

getConfidenceLevel in interface ConfidenceIntervalForecaster

Returns:

the confidence level.

setConfidenceLevel

public void setConfidenceLevel(double confLevel)

Set the confidence level for confidence intervals.

Specified by:

setConfidenceLevel in interface ConfidenceIntervalForecaster

Parameters:

confLevel - the confidence level to use.

getBaseForecaster

public weka.classifiers.Classifier getBaseForecaster()

Get the base Weka regression scheme being used to make forecasts

Returns:

the base Weka regression scheme

setBaseForecaster

public void setBaseForecaster(weka.classifiers.Classifier f)

Set the base Weka regression scheme to use.

Parameters:

f - the base Weka regression scheme to use for forecasting.

isUsingOverlayData

public boolean isUsingOverlayData()

Returns true if overlay data has been used to train this forecaster, and thus is expected to be supplied for future time steps when making a forecast.

Specified by:

isUsingOverlayData in interface OverlayForecaster

Returns:

true if overlay data is expected.

reset

public void reset()

Reset the forecaster.

Specified by:

reset in interface TSForecaster

buildForecaster

public void buildForecaster(weka.core.Instances insts,
                   PrintStream... progress)
                     throws Exception

Builds a new forecasting model using the supplied training data. The instances in the data are assumed to be sorted in ascending order of time and equally spaced in time. Some methods may not need to implement this method and may instead do their work in the primeForecaster method.

Specified by:

buildForecaster in interface TSForecaster

Specified by:

buildForecaster in class AbstractForecaster

Parameters:

insts - the training instances.

progress - an optional varargs parameter supplying progress objects to report/log to

Throws:

Exception - if the model can't be constructed for some reason.

toString

public String toString()

Overrides:

toString in class Object

primeForecaster

public void primeForecaster(weka.core.Instances insts)
                     throws Exception

Supply the (potentially) trained model with enough historical data, up to and including the current time point, in order to produce a forecast. Instances are assumed to be sorted in ascending order of time and equally spaced in time.

Specified by:

primeForecaster in interface TSForecaster

Specified by:

primeForecaster in class AbstractForecaster

Parameters:

insts - the instances to prime the model with

Throws:

Exception - if the model can't be primed for some reason.

primeForecasterIncremental

public void primeForecasterIncremental(weka.core.Instance inst)
                                throws Exception

Update the priming information incrementally, i.e. one instance at a time. To indicate the start of a new batch of priming data an empty set of instances must be passed to TSForecaster.primeForecaster() before the first call to primeForecasterIncremental()

Specified by:

primeForecasterIncremental in interface IncrementallyPrimeable

Parameters:

inst - the instance to prime with.

Throws:

Exception - if something goes wrong.

forecast

public List<List<weka.classifiers.evaluation.NumericPrediction>> forecast(int numSteps,
                                                                 PrintStream... progress)
                                                                   throws Exception

Produce a forecast for the target field(s). Assumes that the model has been built and/or primed so that a forecast can be generated.

Specified by:

forecast in interface TSForecaster

Specified by:

forecast in class AbstractForecaster

Parameters:

numSteps - number of forecasted values to produce for each target. E.g. a value of 5 would produce a prediction for t+1, t+2, ..., t+5. if no overlay data has been used during training)

progress - an optional varargs parameter supplying progress objects to report/log to

Returns:

a List of Lists (one for each step) of forecasted values for each target

Throws:

Exception - if the forecast can't be produced for some reason.

forecast

public List<List<weka.classifiers.evaluation.NumericPrediction>> forecast(int numSteps,
                                                                 weka.core.Instances overlay,
                                                                 PrintStream... progress)
                                                                   throws Exception

Produce a forecast for the target field(s). Assumes that the model has been built and/or primed so that a forecast can be generated.

Specified by:

forecast in interface OverlayForecaster

Parameters:

numSteps - number of forecasted values to produce for each target. E.g. a value of 5 would produce a prediction for t+1, t+2, ..., t+5.

overlay - optional overlay data for the period to be forecasted (may be null if no overlay data has been used during training)

progress - an optional varargs parameter supplying progress objects to report/log to

Returns:

a List of Lists (one for each step) of forecasted values for each target

Throws:

Exception - if the forecast can't be produced for some reason.