weka.filters.supervised.attribute

Class TSLagMaker.PeriodicityHandler

java.lang.Object

weka.filters.supervised.attribute.TSLagMaker.PeriodicityHandler

All Implemented Interfaces:

Serializable

Enclosing class:

TSLagMaker

public static class TSLagMaker.PeriodicityHandler
extends Object
implements Serializable

Helper class to manage time stamp manipulation with respect to various periodicities. Has a routine to remap the time stamp, which is useful for date time stamps. Since dates are just manipulated internally as the number of milliseconds elapsed since the epoch, and any global trend modelling in regression functions results in enormous coefficients for this variable - remapping to a more reasonable scale prevents this. It also makes it easier to handle the case where there are time periods that shouldn't be considered as a time unit increment, e.g. weekends and public holidays for financial trading data. These "holes" in the data can be accomodated by accumulating a negative offset for the remapped date when a particular data/time occurs in a user-specified "skip" list.

Author:

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

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
TSLagMaker.PeriodicityHandler()

Method Summary

Methods

Modifier and Type	Method and Description
boolean	dateInSkipList(Date toCheck) Checks to see if the supplied date is in the list of time units to skip (i.e.
double	deltaTime() Get the delta time of the periodicity being managed
long	getDateTimeStampFinal() Get the last date timestamp value in the batch training data
long	getDateTimeStampInitial() Get the first date time stamp value in the batch training data
TSLagMaker.Periodicity	getPeriodicity() Get periodicity being managed
boolean	isDateBased() Returns true if the periodicity being managed is date timestamp-based
weka.core.Instance	remapDateTimeStamp(weka.core.Instance inst, weka.core.Instance previous, String timeStampName) Remaps a date timestamp to an integer starting (from the first time stamp seen in the data) at 0.
void	setDateTimeStampFinal(long tsfinal) Set the last date timestamp value in the batch training data
void	setDateTimeStampInitial(long tsbase) Set the first date time stamp value in the batch training data
void	setDeltaTime(double deltaTime) Set the delta time for the periodicity being managed
void	setIsDateBased(boolean isDateBased) Set whether the periodicity being managed is date timestamp-based
void	setPeriodicity(TSLagMaker.Periodicity p) Set periodicity to manage
void	setSkipList(String aList, String dateFormat) Set a list of skip entries

Methods inherited from class java.lang.Object

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

Constructor Detail

TSLagMaker.PeriodicityHandler

public TSLagMaker.PeriodicityHandler()

Method Detail

getPeriodicity

public TSLagMaker.Periodicity getPeriodicity()

Get periodicity being managed

Returns:

the periodicity being managed

setPeriodicity

public void setPeriodicity(TSLagMaker.Periodicity p)

Set periodicity to manage

Parameters:

p - the periodicity to manage

setSkipList

public void setSkipList(String aList,
               String dateFormat)
                 throws Exception

Set a list of skip entries

Parameters:

aList - a comma separated list of date-based entries. May include strings such as 'sat' or 'june', specific dates (with optional format string) such as '2011-08-22@yyyy-MM-dd' or integers (which get interpreted differently depending on the periodicity)

dateFormat - a default date format to use for parsing dates

Throws:

Exception - if an entry in the list is unparsable or unrecognized

deltaTime

public double deltaTime()

Get the delta time of the periodicity being managed

Returns:

the delta time

setDeltaTime

public void setDeltaTime(double deltaTime)

Set the delta time for the periodicity being managed

Parameters:

deltaTime - the delta time to use

getDateTimeStampInitial

public long getDateTimeStampInitial()
                             throws Exception

Get the first date time stamp value in the batch training data

Returns:

the first date time stamp value in the batch training data

Throws:

Exception - if the periodicity being managed is not date timestamp-based

setDateTimeStampInitial

public void setDateTimeStampInitial(long tsbase)

Set the first date time stamp value in the batch training data

Parameters:

tsbase - the first date time stamp value in the batch training data as a long (num milliseconds since epoch)

getDateTimeStampFinal

public long getDateTimeStampFinal()
                           throws Exception

Get the last date timestamp value in the batch training data

Returns:

the last date timestamp value in the batch training data

Throws:

Exception - if the periodicity being managed is not date timestamp-based

setDateTimeStampFinal

public void setDateTimeStampFinal(long tsfinal)

Set the last date timestamp value in the batch training data

Parameters:

tsfinal - the last date timestamp value in the batch training data as a long (num milliseconds since the epoch).

setIsDateBased

public void setIsDateBased(boolean isDateBased)

Set whether the periodicity being managed is date timestamp-based

Parameters:

isDateBased - true if the periodicity being managed is date timestamp-based

isDateBased

public boolean isDateBased()

Returns true if the periodicity being managed is date timestamp-based

Returns:

true if the periodicity being managed is date timestamp-based

dateInSkipList

public boolean dateInSkipList(Date toCheck)

Checks to see if the supplied date is in the list of time units to skip (i.e. should not be considered as a time increment).

Parameters:

toCheck - the date to check

Returns:

true if the date is in the skip list

remapDateTimeStamp

public weka.core.Instance remapDateTimeStamp(weka.core.Instance inst,
                                    weka.core.Instance previous,
                                    String timeStampName)
                                      throws Exception

Remaps a date timestamp to an integer starting (from the first time stamp seen in the data) at 0. This is makes any coefficients produced by a regression model for the timestamp (global trend modelling) of reasonable scale. It is also useful for dealing with time units that shouldn't be considered an increment as a negative adjustment can be accumulated for these.

Parameters:

inst - the instance containing a date timestamp to be remapped

previous - the immediately previous instance in the sequence (may be null).

timeStampName - the name of the timestamp attribute

Returns:

Throws:

Exception - if an error occurs