Package com.imsl.datamining.neural

Class TimeSeriesFilter

java.lang.Object
com.imsl.datamining.neural.TimeSeriesFilter
All Implemented Interfaces:
Serializable
public class TimeSeriesFilter extends Object implements Serializable
Converts time series data to a lagged format used as input to a neural network.

Class TimeSeriesFilter can be used to operate on a data matrix and lags every column to form a new data matrix. Using the method computeLags, each column of the input matrix, x, is transformed into (nLags+1) columns by creating a column for \(\text{lags} = 0, 1, \ldots \text{nLags}\).

The output data array, z, can be symbolically represented as: $$z = | x(0) : x(1) : x(2) : \ldots : x(\text{nLags}-1) |,$$ where x(i) is a lagged column of the incoming data matrix, x.

Consider, an example in which x has five rows and two columns with all variables continuous input attributes. Using nObs and nVar to represent the number of rows and columns in x, let $$x = \left[ \begin{array}{cc} 1 & 6 \\ 2 & 7 \\ 3 & 8 \\ 4 & 9 \\ 5 & 10 \\ \end{array} \right]$$ If nLags=1, then the number of columns in z[][] is nVar*(nLags+1)=2*2=4, and the number of rows is (nObs-nLags)=5-1=4: $$z = \left[ \begin{array}{cccc} 1 & 6 & 2 & 7\\ 2 & 7 & 3 & 8 \\ 3 & 8 & 4 & 9 \\ 4 & 9 & 5 & 10 \\ \end{array} \right]$$ If nLags=2, then the number of rows in z will be (nObs-nLags)=(5-2)=3 and the number of columns will be nVar*(nLags+1)=2*3=6: $$z = \left[ \begin{array}{cccccc} 1 & 6 & 2 & 7 & 3 & 8\\ 2 & 7 & 3 & 8 & 4 & 9 \\ 3 & 8 & 4 & 9 & 5 & 10 \\ \end{array} \right]$$

See Also:

  • Constructor Details

    • TimeSeriesFilter

      public TimeSeriesFilter()
      Constructor for TimeSeriesClassFilter.

  • Method Details

    • computeLags

      public double[][] computeLags(int nLags, double[][] x)
      Lags time series data to a format used for input to a neural network.
      Parameters:
      nLags - An int containing the requested number of lags. nLags must be greater than 0.
      x - A double matrix, nObs by nVar, containing the time series data to be lagged. It is assumed that x is sorted in descending chronological order.
      Returns:
      A double matrix with (nObs-nLags) rows and (nVar(nLags+1)) columns. The columns 0 through (nVar-1) contain the columns of x. The next nVar columns contain the first lag of the columns in x, etc.