/**
* Utility to facilitate fitting data plotted in JFreeChart Provide data in JFReeChart format
* (XYSeries), and retrieve univariate function parameters that best fit (using least squares) the
* data. All data points will be weighted equally.
*
* <p>TODO: investigate whether weighting (possibly automatic weighting) can improve accuracy
*
* @param data xy series in JFReeChart format
* @param type one of the Fitter.FunctionType predefined functions
* @param guess initial guess for the fit. The number and meaning of these parameters depends on
* the FunctionType. Implemented: Gaussian: 0: Normalization, 1: Mean 2: Sigma
* @return array with parameters, whose meaning depends on the FunctionType. Use the function
* getXYSeries to retrieve the XYDataset predicted by this fit
*/
public static double[] fit(XYSeries data, FunctionType type, double[] guess) {
if (type == FunctionType.NoFit) {
return null;
}
// create the commons math data object from the JFreeChart data object
final WeightedObservedPoints obs = new WeightedObservedPoints();
for (int i = 0; i < data.getItemCount(); i++) {
obs.add(1.0, data.getX(i).doubleValue(), data.getY(i).doubleValue());
}
double[] result = null;
switch (type) {
case Pol1:
final PolynomialCurveFitter fitter1 = PolynomialCurveFitter.create(1);
result = fitter1.fit(obs.toList());
break;
case Pol2:
final PolynomialCurveFitter fitter2 = PolynomialCurveFitter.create(2);
result = fitter2.fit(obs.toList());
break;
case Pol3:
final PolynomialCurveFitter fitter3 = PolynomialCurveFitter.create(3);
result = fitter3.fit(obs.toList());
break;
case Gaussian:
final GaussianWithOffsetCurveFitter gf = GaussianWithOffsetCurveFitter.create();
if (guess != null) {
gf.withStartPoint(guess);
}
result = gf.fit(obs.toList());
}
return result;
}
/**
* Approximate an already fitted model to polynomial only terms.
*
* <p>This method is mainly used in order to combine the large amplitude long periods with the
* secular part as a new approximate polynomial model over some time range. This should be used
* rather than simply extracting the polynomial coefficients from {@link #getFittedParameters()}
* when some periodic terms amplitudes are large (for example Sun resonance effects on local solar
* time in sun synchronous orbits). In theses cases, the pure polynomial secular part in the
* coefficients may be far from the mean model.
*
* @param combinedDegree desired degree for the combined polynomial
* @param combinedReference desired reference date for the combined polynomial
* @param meanDegree degree of polynomial secular part to consider
* @param meanHarmonics number of harmonics terms to consider
* @param start start date of the approximation time range
* @param end end date of the approximation time range
* @param step sampling step
* @return coefficients of the approximate polynomial (in increasing degree order), using the user
* provided reference date
*/
public double[] approximateAsPolynomialOnly(
final int combinedDegree,
final AbsoluteDate combinedReference,
final int meanDegree,
final int meanHarmonics,
final AbsoluteDate start,
final AbsoluteDate end,
final double step) {
final List<WeightedObservedPoint> points = new ArrayList<WeightedObservedPoint>();
for (AbsoluteDate date = start; date.compareTo(end) < 0; date = date.shiftedBy(step)) {
points.add(
new WeightedObservedPoint(
1.0,
date.durationFrom(combinedReference),
meanValue(date, meanDegree, meanHarmonics)));
}
return PolynomialCurveFitter.create(combinedDegree).fit(points);
}
public static void main(String[] args) {
//////////////////////////
// The values in following 4 lines should be user input
int startPos = 200;
int endPos = 1000;
int totalNumPixel = 1044;
double threshhold = 0.0001;
/////////////////////////
int numPixel = endPos - startPos;
double wavelength[] = new double[totalNumPixel];
double photonCount[] = new double[totalNumPixel];
double SpecNoBk[] = new double[numPixel];
double ThisSpectrum[] = new double[numPixel];
double ThisXaxis[] = new double[numPixel];
double Po[] = new double[numPixel];
double Re[] = new double[numPixel];
double P[] = new double[6];
double Re2[] = new double[numPixel - 1];
double mySUM[] = new double[numPixel];
int ind[];
double DEV;
double prevDEV;
Connection connection = null;
Statement stmt = null;
String pattern = "##.##";
try {
Scanner in = new Scanner(new FileReader(args[0]));
int i = 0;
while (in.hasNextDouble()) {
wavelength[i] = in.nextDouble();
photonCount[i] = in.nextDouble();
++i;
}
} catch (FileNotFoundException e) {
e.printStackTrace();
}
ThisSpectrum = Arrays.copyOfRange(photonCount, startPos, endPos);
ThisXaxis = Arrays.copyOfRange(wavelength, startPos, endPos);
final WeightedObservedPoints obs = new WeightedObservedPoints();
for (int i = 0; i < numPixel; i++) {
obs.add(ThisXaxis[i], ThisSpectrum[i]);
}
final PolynomialCurveFitter fitter = PolynomialCurveFitter.create(5);
P = fitter.fit(obs.toList());
Polyval pVal = new Polyval(P, ThisXaxis, numPixel);
Po = pVal.evl();
for (int i = 0; i < numPixel; i++) {
Re[i] = ThisSpectrum[i] - Po[i];
}
for (int i = 0; i < numPixel - 1; i++) {
Re2[i] = Re[i + 1] - Re[i];
}
DEV = Math.sqrt(StatUtils.populationVariance(Re2, 0, Re2.length));
for (int i = 0; i < numPixel; i++) {
mySUM[i] = Po[i] + DEV;
}
int jj = 0; // jj is the length of points to be removed
for (int i = 0; i < numPixel; i++) {
if (ThisSpectrum[i] > mySUM[i]) {
jj++;
;
}
}
ind = new int[jj];
int jjj = 0;
for (int i = 0; i < numPixel; i++) {
if (ThisSpectrum[i] > mySUM[i]) {
ind[jjj] = i;
jjj++;
}
}
int indKeepLength = numPixel - ind.length;
int indKeep[] = new int[indKeepLength];
int k = 0;
for (int i = 0; i < numPixel; i++) {
if (!ArrayUtils.contains(ind, i)) {
indKeep[k] = i;
k++;
}
}
double ThisSpectrumKeep[] = new double[indKeepLength];
double ThisXaxisKeep[] = new double[indKeepLength];
double PoKeep[] = new double[indKeepLength];
double ReKeep[] = new double[indKeepLength];
double Re2Keep[] = new double[indKeepLength - 1];
double mySUMKeep[] = new double[indKeepLength];
for (int i = 0; i < indKeepLength; i++) {
ThisSpectrumKeep[i] = ThisSpectrum[indKeep[i]];
ThisXaxisKeep[i] = ThisXaxis[indKeep[i]];
}
prevDEV = DEV;
// at the point, ThisSpectrum and ThisXaxis should have reduced size
final WeightedObservedPoints obs1 = new WeightedObservedPoints();
for (int i = 0; i < indKeepLength; i++) {
obs1.add(ThisXaxisKeep[i], ThisSpectrumKeep[i]);
}
while (true) {
final PolynomialCurveFitter fitter1 = PolynomialCurveFitter.create(5);
P = fitter1.fit(obs1.toList());
Polyval pVal1 = new Polyval(P, ThisXaxisKeep, indKeepLength);
PoKeep = pVal1.evl();
for (int i = 0; i < indKeepLength; i++) {
ReKeep[i] = ThisSpectrumKeep[i] - PoKeep[i];
}
for (int i = 0; i < indKeepLength - 1; i++) {
Re2Keep[i] = ReKeep[i + 1] - ReKeep[i];
}
DEV = Math.sqrt(StatUtils.populationVariance(Re2Keep, 0, Re2Keep.length));
for (int i = 0; i < indKeepLength; i++) {
mySUMKeep[i] = PoKeep[i] + DEV;
}
for (int i = 0; i < indKeepLength; i++) {
if (ThisSpectrumKeep[i] > mySUMKeep[i]) ThisSpectrumKeep[i] = mySUMKeep[i];
}
if ((Math.abs(DEV - prevDEV) / DEV) < threshhold) break;
prevDEV = DEV;
obs1.clear();
for (int i = 0; i < indKeepLength; i++) {
obs1.add(ThisXaxisKeep[i], ThisSpectrumKeep[i]);
}
}
Polyval pVal2 = new Polyval(P, ThisXaxis, numPixel);
double FLbk[] = pVal2.evl();
for (int i = 0; i < ThisXaxis.length; i++) {
SpecNoBk[i] = ThisSpectrum[i] - FLbk[i];
}
// the write-to-file part is only for testing purpose, ThisXaxis and SpecNoBk are two outputs
try {
FileWriter fr = new FileWriter(args[1]);
BufferedWriter br = new BufferedWriter(fr);
PrintWriter out = new PrintWriter(br);
DecimalFormat df = new DecimalFormat(pattern);
for (int j = 0; j < ThisXaxis.length; j++) {
if (Double.toString(wavelength[j]) != null) out.write(ThisXaxis[j] + "\t" + SpecNoBk[j]);
out.write("\r\n");
}
out.close();
} catch (IOException e) {
System.out.println(e);
}
}
