Class Spline3

java.lang.Object
org.apache.jmeter.visualizers.Spline3

public class Spline3 extends Object
This class implements the representation of an interpolated Spline curve.

The curve described by such an object interpolates an arbitrary number of fixed points called nodes. The distance between two nodes should currently be constant. This is about to change in a later version but it can last a while as it's not really needed. Nevertheless, if you need the feature, just write me a note and I'll write it asap.

The interpolated Spline curve can't be described by an polynomial analytic equation, the degree of which would be as high as the number of nodes, which would cause extreme oscillations of the curve on the edges.

The solution is to split the curve accross a lot of little intervals : an interval starts at one node and ends at the next one. Then, the interpolation is done on each interval, according to the following conditions :

  1. the interpolated curve is degree 3 : it's a cubic curve ;
  2. the interpolated curve contains the two points delimiting the interval. This condition obviously implies the curve is continuous ;
  3. the interpolated curve has a smooth slope : the curvature has to be the same on the left and the right sides of each node ;
  4. the curvature of the global curve is 0 at both edges.
Every part of the global curve is represented by a cubic (degree-3) polynomial, the coefficients of which have to be computed in order to meet the above conditions.

This leads to a n-unknow n-equation system to resolve. One can resolve an equation system by several manners ; this class uses the Jacobi iterative method, particularly well adapted to this situation, as the diagonal of the system matrix is strong compared to the other elements. This implies the algorithm always converges ! This is not the case of the Gauss-Seidel algorithm, which is quite faster (it uses intermediate results of each iteration to speed up the convergence) but it doesn't converge in all the cases or it converges to a wrong value. This is not acceptable and that's why the Jacobi method is safer. Anyway, the gain of speed is about a factor of 3 but, for a 100x100 system, it means 10 ms instead of 30 ms, which is a pretty good reason not to explore the question any further :)

Here is a little piece of code showing how to use this class :

  // ...
  float[] nodes = {3F, 2F, 4F, 1F, 2.5F, 5F, 3F};
  Spline3 curve = new Spline3(nodes);
  // ...
  public void paint(Graphics g) {
    int[] plot = curve.getPlots();
    for (int i = 1; i < n; i++) {
      g.drawLine(i - 1, plot[i - 1], i, plot[i]);
    }
  }
  // ...
 
  • Field Details

    • _coefficients

      protected float[][] _coefficients
    • _A

      protected float[][] _A
    • _B

      protected float[] _B
    • _r

      protected float[] _r
    • _rS

      protected float[] _rS
    • _m

      protected int _m
    • _n

      protected int _n
    • DEFAULT_PRECISION

      protected static final float DEFAULT_PRECISION
      See Also:
    • DEFAULT_MAX_ITERATIONS

      protected static final int DEFAULT_MAX_ITERATIONS
      See Also:
    • _minPrecision

      protected float _minPrecision
    • _maxIterations

      protected int _maxIterations
  • Constructor Details

    • Spline3

      public Spline3(float[] r)
      Creates a new Spline curve by calculating the coefficients of each part of the curve, i.e. by resolving the equation system implied by the interpolation condition on every interval.
      Parameters:
      r - an array of float containing the vertical coordinates of the nodes to interpolate ; the vertical coordinates start at 0 and are equidistant with a step of 1.
  • Method Details

    • interpolation

      protected void interpolation()
      Computes the coefficients of the Spline interpolated curve, on each interval. The matrix system to resolve is AX=B
    • jacobi

      protected void jacobi()
      Resolves the equation system by a Jacobi algorithm. The use of the slower Jacobi algorithm instead of Gauss-Seidel is choosen here because Jacobi is assured of to be convergent for this particular equation system, as the system matrix has a strong diagonal.
    • converge

      protected boolean converge()
      Test if the Jacobi resolution of the equation system converges. It's OK if A has a strong diagonal.
      Returns:
      true if equation system converges
    • precision

      protected float precision(float[] oldX, float[] newX)
      Computes the current precision reached.
      Parameters:
      oldX - old values
      newX - new values
      Returns:
      indicator of how different the old and new values are (always zero or greater, the nearer to zero the more similar)
    • value

      public float value(float t)
      Computes a (vertical) Y-axis value of the global curve.
      Parameters:
      t - abscissa
      Returns:
      computed ordinate
    • debugCheck

      public void debugCheck()
      Manual check of the curve at the interpolated points.
    • getPlots

      public int[] getPlots(int width, int height)
      Computes drawable plots from the curve for a given draw space. The values returned are drawable vertically and from the bottom of a Panel.
      Parameters:
      width - width within the plots have to be computed
      height - height within the plots are expected to be drawed
      Returns:
      drawable plots within the limits defined, in an array of int (as many int as the value of the width parameter)
    • setPrecision

      public void setPrecision(float precision)
    • getPrecision

      public float getPrecision()
    • setToDefaultPrecision

      public void setToDefaultPrecision()
    • getDefaultPrecision

      public float getDefaultPrecision()
    • setMaxIterations

      public void setMaxIterations(int iterations)
    • getMaxIterations

      public int getMaxIterations()
    • setToDefaultMaxIterations

      public void setToDefaultMaxIterations()
    • getDefaultMaxIterations

      public int getDefaultMaxIterations()