JMSLTM Numerical Library 5.0.1

com.imsl.datamining.neural
Class TimeSeriesClassFilter

java.lang.Object
  extended by com.imsl.datamining.neural.TimeSeriesClassFilter
All Implemented Interfaces:
Serializable

public class TimeSeriesClassFilter
extends Object
implements Serializable

Converts time series data contained within nominal categories to a lagged format for processing by a neural network. Lagging is done within the nominal categories associated with the time series.

Class TimeSeriesClassFilter can be used with a data array, x[] to compute a new data array, z[][], containing lagged columns of x[].

When using the method computeLags, the output array, z[][] of lagged columns, can be symbolically represented as:

z = |x(0) : x(1) : x(2) : ldots : x(nLags-1) |,

where x(i) is a lagged column of the incoming data array x, and nLags is the number of computed lags. The lag associated with x(i) is equal to the value in lag[i], and lagging is done within the nominal categories given in iClass[]. This requires the time series data in x[] be sorted in time order within each category iClass.

Consider an example in which the number of observations in x[] is 10. There are two lags requested in lag[]. If

x^T = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10},

iClass^T = {1, 1, 1, 1, 1, 1, 1, 1, 1, 1},

and

lag^T = {0, 2}

then, all the time series data fall into a single category, i.e. nClasses = 1, and z would contain 2 columns and 10 rows. The first column reproduces the values in x[] because lags[0]=0, and the second column is the 2nd lag because lags[1]=2.

z = left[ begin{array}{cc}
1 & 3 \
2 & 4 \
3 & 5 \
4 & 6 \
5 & 7 \
6 & 8 \
7 & 9 \
8 & 10 \
9 & NaN \
10 & NaN
end{array}  right]

On the other hand, if the data were organized into two classes with

iClass^T = {1, 1, 1, 1, 1, 2, 2, 2, 2, 2},

then nClasses is 2, and z is still a 2 by 10 matrix, but with the following values:

z = left[ frac{ begin{array} {cc}
1 & 3 \
2 & 4 \
3 & 5 \
4 & NaN \
5 & NaN end{array} }
{ begin{array}{cc}
6 & 8 \
7 & 9 \
8 & 10 \
9 & NaN \
10 & NaN end{array} }
right]

The first 5 rows of z are the lagged columns for the first category, and the last five are the lagged columns for the second category.

See Also:
Example, Serialized Form

Constructor Summary
TimeSeriesClassFilter(int nClasses)
          Constructor for TimeSeriesClassFilter.
 
Method Summary
 double[][] computeLags(int[] lags, int[] iClass, double[] x)
          Computes lags of an array sorted first by class designations and then descending chronological order.
 
Methods inherited from class java.lang.Object
clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait
 

Constructor Detail

TimeSeriesClassFilter

public TimeSeriesClassFilter(int nClasses)
Constructor for TimeSeriesClassFilter.

Parameters:
nClasses - An int specifying the number of nominal categories associated with the time series.
Method Detail

computeLags

public double[][] computeLags(int[] lags,
                              int[] iClass,
                              double[] x)
Computes lags of an array sorted first by class designations and then descending chronological order.

Parameters:
lags - An int array containing the requested lags. Every lag must be non-negative.
iClass - An int array containing class number associated with each element of x, sorted in ascending order. The i-th element is equal to the class associated with the i-th element of x. iClass and x must be the same length.
x - A double array containing the time series data to be lagged. This array is assumed to be sorted first by class designations and then descending chronological order, i.e., most recent observations appear first within a class.
Returns:
A double matrix containing the lagged data. The i-th column of this array is the lagged values of x for a lag equal to lags[i]. The number of rows is equal to the length of x.

JMSLTM Numerical Library 5.0.1

Copyright © 1970-2008 Visual Numerics, Inc.
Built July 8 2008.