예제 #1
0
 protected void out(double[] y) {
   String s, blanks, blanksp, blanksm;
   blanksp = "    ";
   blanksm = "   ";
   int ny = y.length;
   int nlines = ny / 5;
   int nrest = ny % 5;
   for (int i = 0; i < nlines; i++) {
     s = "";
     for (int j = 0; j < 5; j++) {
       int k = i * 5 + j;
       if (y[k] >= 0.) blanks = blanksp;
       else blanks = blanksm;
       s = s + blanks + numForm.format(y[k]);
     }
     df.writeLine(s);
   }
   if (nrest > 0) {
     s = "";
     for (int j = 0; j < nrest; j++) {
       int k = nlines * 5 + j;
       if (y[k] >= 0.) blanks = blanksp;
       else blanks = blanksm;
       s = s + blanks + numForm.format(y[k]);
     }
     df.writeLine(s);
   }
 }
예제 #2
0
 protected void compute() {
   nny = (int) ni[0].parseInput();
   df.writeLine("Number of events is: n = " + nny);
   y = DatanRandom.standardNormal(nny);
   out(y);
   mllg = new MinLogLikeGauss(y);
   x0 = new DatanVector(2);
   x0.setElement(0, 1.);
   x0.setElement(1, 2.);
   MinSim ms = new MinSim(x0, mllg);
   x = ms.getMinPosition();
   fcont = ms.getMinimum();
   df.writeLine("Minimization with MinSim yields x = ");
   df.writeLine(x.toString());
   // covariance matrix
   MinCov mc = new MinCov(x, mllg);
   cx = mc.getCovarianceMatrix(1.);
   df.writeLine("Covariance Matrix cx = ");
   df.writeLine(cx.toString());
   // asymmetric errors
   fcont = fcont + 0.5;
   MinAsy ma = new MinAsy(x, cx, mllg);
   dxasy = ma.getAsymmetricErrors(fcont);
   DatanMatrix as = new DatanMatrix(dxasy);
   df.writeLine("Asymmetic errors:");
   df.writeLine(as.toString());
   df.writeLine("ma.hasConverged() = " + ma.hasConverged());
   plotScatterDiagram();
   plotParameterPlane();
 }
예제 #3
0
 public S3Lsq() {
   String s = "Data fitted to power law (nonlinear case)";
   df = new DatanFrame(getClass().getName(), s);
   // parameter input
   AuxJInputGroup ig = new AuxJInputGroup("Parameter input", "");
   JLabel errorLabel = new JLabel();
   ni[0] = new AuxJNumberInput("n", "number of data points (>=3)", errorLabel);
   ni[0].setProperties("n", true, 3);
   ni[0].setNumberInTextField(20);
   ig.add(ni[0]);
   ni[1] = new AuxJNumberInput("t_0", "first value of controlled variable (>=0)", errorLabel);
   ni[1].setProperties("t_0", false, 0.);
   ni[1].setNumberInTextField(0);
   ig.add(ni[1]);
   ni[2] = new AuxJNumberInput("delta_t", "step width of controlled variable (>=0)", errorLabel);
   ni[2].setProperties("delta_t", false, 0.);
   ni[2].setNumberInTextField(.1);
   ig.add(ni[2]);
   ni[3] = new AuxJNumberInput("x_1", "coefficient in front of exponential term (>0)", errorLabel);
   ni[3].setProperties("x_1", false, .000001);
   ni[3].setNumberInTextField(1.);
   ig.add(ni[3]);
   ni[4] = new AuxJNumberInput("x_2", "exponent (>=0)", errorLabel);
   ni[4].setProperties("x_2", false, 0.);
   ni[4].setNumberInTextField(2.);
   ig.add(ni[4]);
   ni[5] = new AuxJNumberInput("sigma", "size of measurement errors (>0)", errorLabel);
   ni[5].setProperties("sigma", false, 0.00001);
   ni[5].setNumberInTextField(.2);
   ig.add(ni[5]);
   df.add(ig);
   df.add(errorLabel);
   JButton goButton = new JButton("Go");
   GoButtonListener gl = new GoButtonListener();
   goButton.addActionListener(gl);
   df.add(goButton);
   df.pack();
   df.toFront();
 }
예제 #4
0
 public E3Min() {
   numForm = NumberFormat.getNumberInstance(Locale.US);
   numForm.setMaximumFractionDigits(12);
   numForm.setMinimumFractionDigits(12);
   String s =
       "Example demonstrating the use of class MinAsy by fitting a Gaussian to small sample"
           + " and determining the asymmetric errors of parameters by MinAsy";
   df = new DatanFrame(getClass().getName(), s);
   AuxJInputGroup ig = new AuxJInputGroup("Enter number N of events (>= 2, <= 10000)", "");
   JLabel errorLabel = new JLabel();
   ni[0] = new AuxJNumberInput("N", "number of events", errorLabel);
   ig.add(ni[0]);
   ni[0].setProperties("N", true);
   ni[0].setMinimum(2);
   ni[0].setMaximum(10000);
   ni[0].setNumberInTextField(10);
   df.add(ig);
   df.add(errorLabel);
   JButton goButton = new JButton("Go");
   GoButtonListener gl = new GoButtonListener();
   goButton.addActionListener(gl);
   df.add(goButton);
   df.repaint();
 }
예제 #5
0
 protected void compute() {
   n = (int) ni[0].parseInput();
   t0 = ni[1].parseInput();
   deltat = ni[2].parseInput();
   x1 = ni[3].parseInput();
   x2 = ni[4].parseInput();
   sigma = ni[5].parseInput();
   // generate data points
   t = new DatanVector(n);
   y = new DatanVector(n);
   dy = new DatanVector(n);
   rand = DatanRandom.standardNormal(n);
   for (int i = 0; i < n; i++) {
     t.setElement(i, t0 + (double) i * deltat);
     y.setElement(i, x1 * Math.pow(t.getElement(i), x2) + sigma * rand[i]);
     dy.setElement(i, sigma);
   }
   // find 1st approximation of unknowns by method of log-log plot
   tlog = new double[n];
   ylog = new double[n];
   dellog = new double[n];
   int npos = 0;
   for (int i = 0; i < n; i++) {
     if (t.getElement(i) > 0. && y.getElement(i) > 0.) {
       tlog[npos] = Math.log(t.getElement(i));
       ylog[npos] = Math.log(y.getElement(i));
       dellog[npos] = 1.;
       npos++;
     }
   }
   DatanVector vtlog = new DatanVector(npos);
   DatanVector vylog = new DatanVector(npos);
   DatanVector vdellog = new DatanVector(npos);
   for (int j = 0; j < npos; j++) {
     vtlog.setElement(j, tlog[j]);
     vylog.setElement(j, ylog[j]);
     vdellog.setElement(j, dellog[j]);
   }
   LsqPol lp = new LsqPol(vtlog, vylog, vdellog, 2);
   x = lp.getResult();
   x.setElement(0, Math.exp(x.getElement(0)));
   df.writeLine(" x = " + x.toString());
   // perform fit
   int[] list = {1, 1};
   powerlaw = new Powerlaw();
   LsqNon ln = new LsqNon(t, y, dy, x, list, powerlaw);
   x = ln.getResult();
   x1 = x.getElement(0);
   x2 = x.getElement(1);
   cov = ln.getCovarianceMatrix();
   delx1 = Math.sqrt(cov.getElement(0, 0));
   delx2 = Math.sqrt(cov.getElement(1, 1));
   rho = cov.getElement(1, 0) / (delx1 * delx2);
   m = ln.getChiSquare();
   p = 1. - StatFunct.cumulativeChiSquared(m, n - 1);
   // curve of fitted exponential
   xpl = new double[1001];
   ypl = new double[1001];
   dpl = (double) (n - 1) * deltat / 1000.;
   for (int i = 0; i < 1001; i++) {
     xpl[i] = t0 + (double) i * dpl;
     ypl[i] = x1 * Math.pow(xpl[i], x2);
   }
   // prepare data points for plotting
   datx = new double[n];
   daty = new double[n];
   datsx = new double[n];
   datsy = new double[n];
   datrho = new double[n];
   for (int i = 0; i < n; i++) {
     datx[i] = t.getElement(i);
     daty[i] = y.getElement(i);
     datsx[i] = 0.;
     datsy[i] = dy.getElement(i);
     datrho[i] = 0.;
   }
   // display data and fitted curve
   caption =
       "x_1#="
           + String.format(Locale.US, "%5.2f", x1)
           + ", x_2#="
           + String.format(Locale.US, "%5.2f", x2)
           + ", &D@x_1#="
           + String.format(Locale.US, "%5.2f", delx1)
           + ", &D@x_2#="
           + String.format(Locale.US, "%5.2f", delx2)
           + ", &r@="
           + String.format(Locale.US, "%5.2f", rho)
           + ", M="
           + String.format(Locale.US, "%5.2f", m)
           + ", P="
           + String.format(Locale.US, "%6.4f", p);
   new GraphicsWithDataPointsAndPolyline(
       getClass().getName(),
       "",
       xpl,
       ypl,
       1,
       .3,
       datx,
       daty,
       datsx,
       datsy,
       datrho,
       "t",
       "y",
       caption);
 }