/**
* Chi square distribution
*
* @param x Chi^2 value
* @param n Degrees of freedom
* @return P-value associated
*/
private static double ChiSq(double x, int n) {
if (n == 1 & x > 1000) {
return 0;
}
if (x > 1000 | n > 1000) {
double q = ChiSq((x - n) * (x - n) / (2 * n), 1) / 2;
if (x > n) {
return q;
}
{
return 1 - q;
}
}
double p = Math.exp(-0.5 * x);
if ((n % 2) == 1) {
p = p * Math.sqrt(2 * x / Math.PI);
}
double k = n;
while (k >= 2) {
p = p * x / k;
k = k - 2;
}
double t = p;
double a = n;
while (t > 0.0000000001 * p) {
a = a + 2;
t = t * x / a;
p = p + t;
}
return 1 - p;
}
/**
* Claculates the upper bound for the Chi-Squared value of a rule.
*
* @param suppAnte the support for the antecedent of a rule.
* @param suppCons the support for the consequent of a rule.
* @return the Chi-Squared upper bound.
*/
private double calcChiSquaredUpperBound(double suppAnte, double suppCons) {
double term;
// Test support for antecedent and confidence and choose minimum
if (suppAnte < suppCons) term = Math.pow(suppAnte - ((suppAnte * suppCons) / numRecords), 2.0);
else term = Math.pow(suppCons - ((suppAnte * suppCons) / numRecords), 2.0);
// Determine e
double eVlaue = calcWCSeValue(suppAnte, suppCons);
// Rerturn upper bound
return (term * eVlaue * numRecords);
}
/**
* This private method computes the distance between an example in the dataset and a cluster
* centroid. The distance is measure as the square root of the sum of the squares of the
* differences between the example and the cetroid for all the dimensions.
*
* @param a The example in the dataset
* @param b The culster centroid
* @return The distance between a and b as a double precision float value.
*/
private static double distance(double a[], double b[]) {
// Euclid distance between two patterns
double d = 0;
for (int i = 0; i < a.length; i++) d += (a[i] - b[i]) * (a[i] - b[i]);
return (double) Math.sqrt(d);
}
/**
* Computes the ouput of a RBF
*
* @param _input Input vector
* @return The ouput of a RBF
*/
public double evaluationRbf(double[] _input) {
double aux;
aux = RBFUtils.euclidean(_input, centre);
aux *= aux;
aux /= (2.0 * radius * radius);
return (Math.exp(-aux));
}
/**
* It converts double to char by the gray's code
*
* @param ds the vector which is goint to change
* @param length the size of the vector
* @return the changed vector
*/
private char[] StringRep(double[] ds, int length) {
int i;
double n;
int pos;
double INCREMENTO;
char[] Cad_sal;
Cad_sal = new char[Genes * BITS_GEN + 1];
if (flag == 1) {
tmpstring = new char[Genes * BITS_GEN];
flag = 0;
}
pos = 0;
for (i = 0; i < length; i++) {
INCREMENTO = (Gene[i].max() - Gene[i].min()) / (Math.pow(2.0, (double) BITS_GEN) - 1.0);
n = (((ds[i] - Gene[i].min()) / INCREMENTO) + 0.5);
tmpstring = F.Itoc((int) n, BITS_GEN);
F.Gray(tmpstring, Cad_sal, BITS_GEN, pos);
pos += BITS_GEN;
}
return Cad_sal;
}
/**
* Computes the euclidean distance between a neuron and a vector
*
* @param v A vector
* @return A double with the euclidean distance
*/
public double euclideaDist(double[] v) {
int i;
double aux = 0;
for (i = 0; i < nInput; i++) aux += (v[i] - centre[i]) * (v[i] - centre[i]);
return (Math.sqrt(aux));
}
/**
* Fisher distribution
*
* @param f Fisher value
* @param n1 N1 value
* @param n2 N2 value
* @return P-value associated
*/
private static double FishF(double f, int n1, int n2) {
double x = n2 / (n1 * f + n2);
if ((n1 % 2) == 0) {
return StatCom(1 - x, n2, n1 + n2 - 4, n2 - 2) * Math.pow(x, n2 / 2.0);
}
if ((n2 % 2) == 0) {
return 1 - StatCom(x, n1, n1 + n2 - 4, n1 - 2) * Math.pow(1 - x, n1 / 2.0);
}
double th = Math.atan(Math.sqrt(n1 * f / (1.0 * n2)));
double a = th / (Math.PI / 2.0);
double sth = Math.sin(th);
double cth = Math.cos(th);
if (n2 > 1) {
a = a + sth * cth * StatCom(cth * cth, 2, n2 - 3, -1) / (Math.PI / 2.0);
}
if (n1 == 1) {
return 1 - a;
}
double c =
4 * StatCom(sth * sth, n2 + 1, n1 + n2 - 4, n2 - 2) * sth * Math.pow(cth, n2) / Math.PI;
if (n2 == 1) {
return 1 - a + c / 2.0;
}
int k = 2;
while (k <= (n2 - 1) / 2.0) {
c = c * k / (k - .5);
k = k + 1;
}
return 1 - a + c;
}
/**
* Calculates the Chi squared values and returns their sum.
*
* @return the sum of the Chi Squared values.
*/
private double calcChiSquaredValue() {
double sumChiSquaredValues = 0.0;
for (int index = 0; index < obsValues.length; index++) {
double chiValue = Math.pow((obsValues[index] - expValues[index]), 2.0) / expValues[index];
sumChiSquaredValues = sumChiSquaredValues + chiValue;
}
// Return
return (sumChiSquaredValues);
}
private double computeAdjustedResidual(int data[][], int classData[], Rule regla, int clase) {
double suma = 0;
int i;
for (i = 0; i < regla.getRule().length; i++) {
suma +=
computeStandarizedResidual(data, classData, regla.getiCondition(i), clase)
/ Math.sqrt(
computeMaximumLikelohoodEstimate(data, classData, regla.getiCondition(i), clase));
}
return suma;
}
/**
* It calculate the matching degree between the antecedent of the rule and a given example
*
* @param indiv Individual The individual representing a fuzzy rule
* @param ejemplo double [] A given example
* @return double The matching degree between the example and the antecedent of the rule
*/
double Matching_degree(Individual indiv, double[] ejemplo) {
int i, sig;
double result, suma, numerador, denominador, sigma, ancho_intervalo;
suma = 0.0;
for (i = 0; i < entradas; i++) {
ancho_intervalo = train.getMax(i) - train.getMin(i);
sigma = -1.0;
sig = indiv.antecedente[i].sigma;
switch (sig) {
case 1:
sigma = 0.3;
break;
case 2:
sigma = 0.4;
break;
case 3:
sigma = 0.5;
break;
case 4:
sigma = 0.6;
break;
}
sigma *= ancho_intervalo;
numerador = Math.pow((ejemplo[i] - indiv.antecedente[i].m), 2.0);
denominador = Math.pow(sigma, 2.0);
suma += (numerador / denominador);
}
suma *= -1.0;
result = Math.exp(suma);
return (result);
}
private void interpola(
double ra[],
double rb[],
int na[],
int nb[],
boolean ma[],
boolean mb[],
double resS[],
double resR[],
int resN[],
boolean resM[]) {
int i;
double diff;
double gap;
int suerte;
for (i = 0; i < ra.length; i++) {
if (ma[i] == true && mb[i] == true) {
resM[i] = true;
resS[i] = 0;
} else {
resM[i] = false;
if (entradas[i].getType() == Attribute.REAL) {
diff = rb[i] - ra[i];
gap = Randomize.Rand();
resR[i] = ra[i] + gap * diff;
resS[i] =
(ra[i] + entradas[i].getMinAttribute())
/ (entradas[i].getMaxAttribute() - entradas[i].getMinAttribute());
} else if (entradas[i].getType() == Attribute.INTEGER) {
diff = rb[i] - ra[i];
gap = Randomize.Rand();
resR[i] = Math.round(ra[i] + gap * diff);
resS[i] =
(ra[i] + entradas[i].getMinAttribute())
/ (entradas[i].getMaxAttribute() - entradas[i].getMinAttribute());
} else {
suerte = Randomize.Randint(0, 2);
if (suerte == 0) {
resN[i] = na[i];
} else {
resN[i] = nb[i];
}
resS[i] = (double) resN[i] / (double) (entradas[i].getNominalValuesList().size() - 1);
}
}
}
}
/**
* It Evaluates the performance of the fuzzy system. The Mean Square Error (MSE) by training is
* used
*/
public double Evaluate_fuzzy_system() {
int i;
double result, suma, fuerza;
suma = 0.0;
for (i = 0; i < train.getnData(); i++) {
fuerza = Output_fuzzy_system(train.getExample(i));
suma += Math.pow(train.getOutputAsReal(i) - fuerza, 2.0);
}
result = suma / train.getnData();
/* We want to have a maximization problem so, we invert the error */
if (result != 0.0) {
result = 1.0 / result;
}
return (result);
}
/**
* It Evaluates the performance of the best evolved fuzzy system on test data. The Mean Square
* Error (MSE) is used
*
* @return double The MSE error in test data
*/
public double Evaluate_best_fuzzy_system_in_test() {
int i;
double result, suma, fuerza;
SistemaDifuso.clear();
for (i = 0; i < Nr; i++) {
Individual indi = new Individual(BestSistemaDifuso.get(i));
SistemaDifuso.add(indi);
}
suma = 0.0;
for (i = 0; i < test.getnData(); i++) {
fuerza = Output_fuzzy_system(test.getExample(i));
suma += Math.pow(test.getOutputAsReal(i) - fuerza, 2.0);
}
result = suma / test.getnData();
return (result);
}
/**
* It calculates PCBLX
*
* @param d it multiplies the module of the difference of the parents
* @param P1 it is a father to cross
* @param P2 it is the other father to do the cross
* @param Hijo1 the son obtained
* @param Hijo2 the other obtained
* @param gens the size of the vector
*/
private void xPC_BLX(
double d, double[] P1, double[] P2, double[] Hijo1, double[] Hijo2, int gens) {
double I, A1, C1;
int i;
for (i = 0; i < gens; i++) {
I = d * Math.abs(P1[i] - P2[i]);
A1 = P1[i] - I;
if (A1 < Gene[i].min()) A1 = Gene[i].min();
C1 = P1[i] + I;
if (C1 > Gene[i].max()) C1 = Gene[i].max();
Hijo1[i] = A1 + Randomize.Rand() * (C1 - A1);
A1 = P2[i] - I;
if (A1 < Gene[i].min()) A1 = Gene[i].min();
C1 = P2[i] + I;
if (C1 > Gene[i].max()) C1 = Gene[i].max();
Hijo2[i] = A1 + Randomize.Rand() * (C1 - A1);
}
}
/**
* This method runs the multiple comparison tests
*
* @param code A value to codify which post-hoc methods apply
* @param results Array with the results of the methods
* @param algorithmName Array with the name of the methods employed
* @return A string with the contents of the test in LaTeX format
*/
private static String runMultiple(double[][] results, String algorithmName[]) {
int i, j, k;
int posicion;
double mean[][];
MultiplePair orden[][];
MultiplePair rank[][];
boolean encontrado;
int ig;
double sum;
boolean visto[];
Vector<Integer> porVisitar;
double Rj[];
double friedman;
double sumatoria = 0;
double termino1, termino2, termino3;
double iman;
boolean vistos[];
int pos, tmp, counter;
String cad;
double maxVal;
double Pi[];
double ALPHAiHolm[];
double ALPHAiShaffer[];
String ordenAlgoritmos[];
double ordenRankings[];
int order[];
double adjustedP[][];
double SE;
boolean parar;
Vector<Integer> indices = new Vector<Integer>();
Vector<Vector<Relation>> exhaustiveI = new Vector<Vector<Relation>>();
boolean[][] cuadro;
double minPi, tmpPi, maxAPi, tmpAPi;
Relation[] parejitas;
Vector<Integer> T;
int Tarray[];
DecimalFormat nf4 = (DecimalFormat) DecimalFormat.getInstance();
nf4.setMaximumFractionDigits(4);
nf4.setMinimumFractionDigits(0);
DecimalFormatSymbols dfs = nf4.getDecimalFormatSymbols();
dfs.setDecimalSeparator('.');
nf4.setDecimalFormatSymbols(dfs);
DecimalFormat nf6 = (DecimalFormat) DecimalFormat.getInstance();
nf6.setMaximumFractionDigits(6);
nf6.setMinimumFractionDigits(0);
nf6.setDecimalFormatSymbols(dfs);
String out = "";
int nDatasets = Configuration.getNDatasets();
Iman = Configuration.isIman();
Nemenyi = Configuration.isNemenyi();
Bonferroni = Configuration.isBonferroni();
Holm = Configuration.isHolm();
Hoch = Configuration.isHochberg();
Hommel = Configuration.isHommel();
Scha = Configuration.isShaffer();
Berg = Configuration.isBergman();
mean = new double[nDatasets][algorithmName.length];
// Maximize performance
if (Configuration.getObjective() == 1) {
/*Compute the average performance per algorithm for each data set*/
for (i = 0; i < nDatasets; i++) {
for (j = 0; j < algorithmName.length; j++) {
mean[i][j] = results[j][i];
}
}
}
// Minimize performance
else {
double maxValue = Double.MIN_VALUE;
/*Compute the average performance per algorithm for each data set*/
for (i = 0; i < nDatasets; i++) {
for (j = 0; j < algorithmName.length; j++) {
if (results[j][i] > maxValue) {
maxValue = results[j][i];
}
mean[i][j] = (-1.0 * results[j][i]);
}
}
for (i = 0; i < nDatasets; i++) {
for (j = 0; j < algorithmName.length; j++) {
mean[i][j] += maxValue;
}
}
}
/*We use the pareja structure to compute and order rankings*/
orden = new MultiplePair[nDatasets][algorithmName.length];
for (i = 0; i < nDatasets; i++) {
for (j = 0; j < algorithmName.length; j++) {
orden[i][j] = new MultiplePair(j, mean[i][j]);
}
Arrays.sort(orden[i]);
}
/*building of the rankings table per algorithms and data sets*/
rank = new MultiplePair[nDatasets][algorithmName.length];
posicion = 0;
for (i = 0; i < nDatasets; i++) {
for (j = 0; j < algorithmName.length; j++) {
encontrado = false;
for (k = 0; k < algorithmName.length && !encontrado; k++) {
if (orden[i][k].indice == j) {
encontrado = true;
posicion = k + 1;
}
}
rank[i][j] = new MultiplePair(posicion, orden[i][posicion - 1].valor);
}
}
/*In the case of having the same performance, the rankings are equal*/
for (i = 0; i < nDatasets; i++) {
visto = new boolean[algorithmName.length];
porVisitar = new Vector<Integer>();
Arrays.fill(visto, false);
for (j = 0; j < algorithmName.length; j++) {
porVisitar.removeAllElements();
sum = rank[i][j].indice;
visto[j] = true;
ig = 1;
for (k = j + 1; k < algorithmName.length; k++) {
if (rank[i][j].valor == rank[i][k].valor && !visto[k]) {
sum += rank[i][k].indice;
ig++;
porVisitar.add(new Integer(k));
visto[k] = true;
}
}
sum /= (double) ig;
rank[i][j].indice = sum;
for (k = 0; k < porVisitar.size(); k++) {
rank[i][((Integer) porVisitar.elementAt(k)).intValue()].indice = sum;
}
}
}
/*compute the average ranking for each algorithm*/
Rj = new double[algorithmName.length];
for (i = 0; i < algorithmName.length; i++) {
Rj[i] = 0;
for (j = 0; j < nDatasets; j++) {
Rj[i] += rank[j][i].indice / ((double) nDatasets);
}
}
/*Print the average ranking per algorithm*/
out += "\n\nAverage ranks obtained by applying the Friedman procedure\n\n";
out +=
"\\begin{table}[!htp]\n"
+ "\\centering\n"
+ "\\begin{tabular}{|c|c|}\\hline\n"
+ "Algorithm&Ranking\\\\\\hline\n";
for (i = 0; i < algorithmName.length; i++) {
out += (String) algorithmName[i] + " & " + nf4.format(Rj[i]) + "\\\\\n";
}
out += "\\hline\n\\end{tabular}\n\\caption{Average Rankings of the algorithms}\n\\end{table}";
/*Compute the Friedman statistic*/
termino1 =
(12 * (double) nDatasets)
/ ((double) algorithmName.length * ((double) algorithmName.length + 1));
termino2 =
(double) algorithmName.length
* ((double) algorithmName.length + 1)
* ((double) algorithmName.length + 1)
/ (4.0);
for (i = 0; i < algorithmName.length; i++) {
sumatoria += Rj[i] * Rj[i];
}
friedman = (sumatoria - termino2) * termino1;
out +=
"\n\nFriedman statistic considering reduction performance (distributed according to chi-square with "
+ (algorithmName.length - 1)
+ " degrees of freedom: "
+ nf6.format(friedman)
+ ".\n\n";
double pFriedman;
pFriedman = ChiSq(friedman, (algorithmName.length - 1));
System.out.print("P-value computed by Friedman Test: " + pFriedman + ".\\newline\n\n");
/*Compute the Iman-Davenport statistic*/
if (Iman) {
iman = ((nDatasets - 1) * friedman) / (nDatasets * (algorithmName.length - 1) - friedman);
out +=
"Iman and Davenport statistic considering reduction performance (distributed according to F-distribution with "
+ (algorithmName.length - 1)
+ " and "
+ (algorithmName.length - 1) * (nDatasets - 1)
+ " degrees of freedom: "
+ nf6.format(iman)
+ ".\n\n";
double pIman;
pIman = FishF(iman, (algorithmName.length - 1), (algorithmName.length - 1) * (nDatasets - 1));
System.out.print("P-value computed by Iman and Daveport Test: " + pIman + ".\\newline\n\n");
}
termino3 =
Math.sqrt(
(double) algorithmName.length
* ((double) algorithmName.length + 1)
/ (6.0 * (double) nDatasets));
out += "\n\n\\pagebreak\n\n";
/** ********** NxN COMPARISON ************* */
out += "\\section{Post hoc comparisons}";
out +=
"\n\nResults achieved on post hoc comparisons for $\\alpha = 0.05$, $\\alpha = 0.10$ and adjusted p-values.\n\n";
/*Compute the unadjusted p_i value for each comparison alpha=0.05*/
Pi = new double[(int) combinatoria(2, algorithmName.length)];
ALPHAiHolm = new double[(int) combinatoria(2, algorithmName.length)];
ALPHAiShaffer = new double[(int) combinatoria(2, algorithmName.length)];
ordenAlgoritmos = new String[(int) combinatoria(2, algorithmName.length)];
ordenRankings = new double[(int) combinatoria(2, algorithmName.length)];
order = new int[(int) combinatoria(2, algorithmName.length)];
parejitas = new Relation[(int) combinatoria(2, algorithmName.length)];
T = new Vector<Integer>();
T = trueHShaffer(algorithmName.length);
Tarray = new int[T.size()];
for (i = 0; i < T.size(); i++) {
Tarray[i] = ((Integer) T.elementAt(i)).intValue();
}
Arrays.sort(Tarray);
SE = termino3;
vistos = new boolean[(int) combinatoria(2, algorithmName.length)];
for (i = 0, k = 0; i < algorithmName.length; i++) {
for (j = i + 1; j < algorithmName.length; j++, k++) {
ordenRankings[k] = Math.abs(Rj[i] - Rj[j]);
ordenAlgoritmos[k] = (String) algorithmName[i] + " vs. " + (String) algorithmName[j];
parejitas[k] = new Relation(i, j);
}
}
Arrays.fill(vistos, false);
for (i = 0; i < ordenRankings.length; i++) {
for (j = 0; vistos[j] == true; j++) ;
pos = j;
maxVal = ordenRankings[j];
for (j = j + 1; j < ordenRankings.length; j++) {
if (vistos[j] == false && ordenRankings[j] > maxVal) {
pos = j;
maxVal = ordenRankings[j];
}
}
vistos[pos] = true;
order[i] = pos;
}
/*Computing the logically related hypotheses tests (Shaffer and Bergmann-Hommel)*/
pos = 0;
tmp = Tarray.length - 1;
for (i = 0; i < order.length; i++) {
Pi[i] = 2 * CDF_Normal.normp((-1) * Math.abs((ordenRankings[order[i]]) / SE));
ALPHAiHolm[i] = 0.05 / ((double) order.length - (double) i);
ALPHAiShaffer[i] = 0.05 / ((double) order.length - (double) Math.max(pos, i));
if (i == pos && Pi[i] <= ALPHAiShaffer[i]) {
tmp--;
pos = (int) combinatoria(2, algorithmName.length) - Tarray[tmp];
}
}
out += "\\subsection{P-values for $\\alpha=0.05$}\n\n";
int count = 4;
if (Holm) {
count++;
}
if (Scha) {
count++;
}
out +=
"\\begin{table}[!htp]\n\\centering\\scriptsize\n"
+ "\\begin{tabular}{"
+ printC(count)
+ "}\n"
+ "$i$&algorithms&$z=(R_0 - R_i)/SE$&$p$";
if (Holm) {
out += "&Holm";
}
if (Scha) {
out += "&Shaffer";
}
out += "\\\\\n\\hline";
for (i = 0; i < order.length; i++) {
out +=
(order.length - i)
+ "&"
+ ordenAlgoritmos[order[i]]
+ "&"
+ nf6.format(Math.abs((ordenRankings[order[i]]) / SE))
+ "&"
+ nf6.format(Pi[i]);
if (Holm) {
out += "&" + nf6.format(ALPHAiHolm[i]);
}
if (Scha) {
out += "&" + nf6.format(ALPHAiShaffer[i]);
}
out += "\\\\\n";
}
out +=
"\\hline\n"
+ "\\end{tabular}\n\\caption{P-values Table for $\\alpha=0.05$}\n"
+ "\\end{table}";
/*Compute the rejected hipotheses for each test*/
if (Nemenyi) {
out +=
"Nemenyi's procedure rejects those hypotheses that have a p-value $\\le"
+ nf6.format(0.05 / (double) (order.length))
+ "$.\n\n";
}
if (Holm) {
parar = false;
for (i = 0; i < order.length && !parar; i++) {
if (Pi[i] > ALPHAiHolm[i]) {
out +=
"Holm's procedure rejects those hypotheses that have a p-value $\\le"
+ nf6.format(ALPHAiHolm[i])
+ "$.\n\n";
parar = true;
}
}
}
if (Scha) {
parar = false;
for (i = 0; i < order.length && !parar; i++) {
if (Pi[i] <= ALPHAiShaffer[i]) {
out +=
"Shaffer's procedure rejects those hypotheses that have a p-value $\\le"
+ nf6.format(ALPHAiShaffer[i])
+ "$.\n\n";
parar = true;
}
}
}
/*For Bergmann-Hommel's procedure, 9 algorithms could suppose intense computation*/
if (algorithmName.length <= MAX_ALGORITHMS) {
for (i = 0; i < algorithmName.length; i++) {
indices.add(new Integer(i));
}
exhaustiveI = obtainExhaustive(indices);
cuadro = new boolean[algorithmName.length][algorithmName.length];
for (i = 0; i < algorithmName.length; i++) {
Arrays.fill(cuadro[i], false);
}
for (i = 0; i < exhaustiveI.size(); i++) {
minPi =
2
* CDF_Normal.normp(
(-1)
* Math.abs(
Rj[
((Relation)
((Vector<Relation>) exhaustiveI.elementAt(i))
.elementAt(0))
.i]
- Rj[
((Relation)
((Vector<Relation>) exhaustiveI.elementAt(i))
.elementAt(0))
.j])
/ SE);
for (j = 1; j < ((Vector<Relation>) exhaustiveI.elementAt(i)).size(); j++) {
tmpPi =
2
* CDF_Normal.normp(
(-1)
* Math.abs(
Rj[
((Relation)
((Vector<Relation>) exhaustiveI.elementAt(i))
.elementAt(j))
.i]
- Rj[
((Relation)
((Vector<Relation>) exhaustiveI.elementAt(i))
.elementAt(j))
.j])
/ SE);
if (tmpPi < minPi) {
minPi = tmpPi;
}
}
if (minPi > (0.05 / ((double) ((Vector<Relation>) exhaustiveI.elementAt(i)).size()))) {
for (j = 0; j < ((Vector<Relation>) exhaustiveI.elementAt(i)).size(); j++) {
cuadro[((Relation) ((Vector<Relation>) exhaustiveI.elementAt(i)).elementAt(j)).i][
((Relation) ((Vector<Relation>) exhaustiveI.elementAt(i)).elementAt(j)).j] =
true;
}
}
}
if (Berg) {
cad = "";
cad += "Bergmann's procedure rejects these hypotheses:\n\n";
cad += "\\begin{itemize}\n\n";
counter = 0;
for (i = 0; i < cuadro.length; i++) {
for (j = i + 1; j < cuadro.length; j++) {
if (cuadro[i][j] == false) {
cad += "\\item " + algorithmName[i] + " vs. " + algorithmName[j] + "\n\n";
counter++;
}
}
}
cad += "\\end{itemize}\n\n";
if (counter > 0) {
out += cad;
} else {
out += "Bergmann's procedure does not reject any hypotheses.\n\n";
}
}
}
out += "\\pagebreak\n\n";
out += "\\subsection{P-values for $\\alpha=0.10$}\n\n";
/*Compute the unadjusted p_i value for each comparison alpha=0.10*/
Pi = new double[(int) combinatoria(2, algorithmName.length)];
ALPHAiHolm = new double[(int) combinatoria(2, algorithmName.length)];
ALPHAiShaffer = new double[(int) combinatoria(2, algorithmName.length)];
ordenAlgoritmos = new String[(int) combinatoria(2, algorithmName.length)];
ordenRankings = new double[(int) combinatoria(2, algorithmName.length)];
order = new int[(int) combinatoria(2, algorithmName.length)];
SE = termino3;
vistos = new boolean[(int) combinatoria(2, algorithmName.length)];
for (i = 0, k = 0; i < algorithmName.length; i++) {
for (j = i + 1; j < algorithmName.length; j++, k++) {
ordenRankings[k] = Math.abs(Rj[i] - Rj[j]);
ordenAlgoritmos[k] = (String) algorithmName[i] + " vs. " + (String) algorithmName[j];
}
}
Arrays.fill(vistos, false);
for (i = 0; i < ordenRankings.length; i++) {
for (j = 0; vistos[j] == true; j++) ;
pos = j;
maxVal = ordenRankings[j];
for (j = j + 1; j < ordenRankings.length; j++) {
if (vistos[j] == false && ordenRankings[j] > maxVal) {
pos = j;
maxVal = ordenRankings[j];
}
}
vistos[pos] = true;
order[i] = pos;
}
/*Computing the logically related hypotheses tests (Shaffer and Bergmann-Hommel)*/
pos = 0;
tmp = Tarray.length - 1;
for (i = 0; i < order.length; i++) {
Pi[i] = 2 * CDF_Normal.normp((-1) * Math.abs((ordenRankings[order[i]]) / SE));
ALPHAiHolm[i] = 0.1 / ((double) order.length - (double) i);
ALPHAiShaffer[i] = 0.1 / ((double) order.length - (double) Math.max(pos, i));
if (i == pos && Pi[i] <= ALPHAiShaffer[i]) {
tmp--;
pos = (int) combinatoria(2, algorithmName.length) - Tarray[tmp];
}
}
count = 4;
if (Holm) {
count++;
}
if (Scha) {
count++;
}
out +=
"\\begin{table}[!htp]\n\\centering\\scriptsize\n"
+ "\\begin{tabular}{"
+ printC(count)
+ "}\n"
+ "$i$&algorithms&$z=(R_0 - R_i)/SE$&$p$";
if (Holm) {
out += "&Holm";
}
if (Scha) {
out += "&Shaffer";
}
out += "\\\\\n\\hline";
for (i = 0; i < order.length; i++) {
out +=
(order.length - i)
+ "&"
+ ordenAlgoritmos[order[i]]
+ "&"
+ nf6.format(Math.abs((ordenRankings[order[i]]) / SE))
+ "&"
+ nf6.format(Pi[i]);
if (Holm) {
out += "&" + nf6.format(ALPHAiHolm[i]);
}
if (Scha) {
out += "&" + nf6.format(ALPHAiShaffer[i]);
}
out += "\\\\\n";
}
out +=
"\\hline\n"
+ "\\end{tabular}\n\\caption{P-values Table for $\\alpha=0.10$}\n"
+ "\\end{table}";
/*Compute the rejected hipotheses for each test*/
if (Nemenyi) {
out +=
"Nemenyi's procedure rejects those hypotheses that have a p-value $\\le"
+ nf6.format(0.10 / (double) (order.length))
+ "$.\n\n";
}
if (Holm) {
parar = false;
for (i = 0; i < order.length && !parar; i++) {
if (Pi[i] > ALPHAiHolm[i]) {
out +=
"Holm's procedure rejects those hypotheses that have a p-value $\\le"
+ nf6.format(ALPHAiHolm[i])
+ "$.\n\n";
parar = true;
}
}
}
if (Scha) {
parar = false;
for (i = 0; i < order.length && !parar; i++) {
if (Pi[i] <= ALPHAiShaffer[i]) {
out +=
"Shaffer's procedure rejects those hypotheses that have a p-value $\\le"
+ nf6.format(ALPHAiShaffer[i])
+ "$.\n\n";
parar = true;
}
}
}
/*For Bergmann-Hommel's procedure, 9 algorithms could suppose intense computation*/
if (algorithmName.length <= MAX_ALGORITHMS) {
indices.removeAllElements();
for (i = 0; i < algorithmName.length; i++) {
indices.add(new Integer(i));
}
exhaustiveI = obtainExhaustive(indices);
cuadro = new boolean[algorithmName.length][algorithmName.length];
for (i = 0; i < algorithmName.length; i++) {
Arrays.fill(cuadro[i], false);
}
for (i = 0; i < exhaustiveI.size(); i++) {
minPi =
2
* CDF_Normal.normp(
(-1)
* Math.abs(
Rj[
((Relation)
((Vector<Relation>) exhaustiveI.elementAt(i))
.elementAt(0))
.i]
- Rj[
((Relation)
((Vector<Relation>) exhaustiveI.elementAt(i))
.elementAt(0))
.j])
/ SE);
for (j = 1; j < ((Vector<Relation>) exhaustiveI.elementAt(i)).size(); j++) {
tmpPi =
2
* CDF_Normal.normp(
(-1)
* Math.abs(
Rj[
((Relation)
((Vector<Relation>) exhaustiveI.elementAt(i))
.elementAt(j))
.i]
- Rj[
((Relation)
((Vector<Relation>) exhaustiveI.elementAt(i))
.elementAt(j))
.j])
/ SE);
if (tmpPi < minPi) {
minPi = tmpPi;
}
}
if (minPi > 0.1 / ((double) ((Vector<Relation>) exhaustiveI.elementAt(i)).size())) {
for (j = 0; j < ((Vector<Relation>) exhaustiveI.elementAt(i)).size(); j++) {
cuadro[((Relation) ((Vector<Relation>) exhaustiveI.elementAt(i)).elementAt(j)).i][
((Relation) ((Vector<Relation>) exhaustiveI.elementAt(i)).elementAt(j)).j] =
true;
}
}
}
if (Berg) {
cad = "";
cad += "Bergmann's procedure rejects these hypotheses:\n\n";
cad += "\\begin{itemize}\n\n";
counter = 0;
for (i = 0; i < cuadro.length; i++) {
for (j = i + 1; j < cuadro.length; j++) {
if (cuadro[i][j] == false) {
cad += "\\item " + algorithmName[i] + " vs. " + algorithmName[j] + "\n\n";
counter++;
}
}
}
cad += "\\end{itemize}\n\n";
if (counter > 0) {
out += cad;
} else {
out += "Bergmann's procedure does not reject any hypotheses.\n\n";
}
}
}
out += "\\pagebreak\n\n";
/** ********** ADJUSTED P-VALUES NxN COMPARISON ************* */
out += "\\subsection{Adjusted p-values}\n\n";
adjustedP = new double[Pi.length][4];
pos = 0;
tmp = Tarray.length - 1;
for (i = 0; i < adjustedP.length; i++) {
adjustedP[i][0] = Pi[i] * (double) (adjustedP.length);
adjustedP[i][1] = Pi[i] * (double) (adjustedP.length - i);
adjustedP[i][2] = Pi[i] * ((double) adjustedP.length - (double) Math.max(pos, i));
if (i == pos) {
tmp--;
pos = (int) combinatoria(2, algorithmName.length) - Tarray[tmp];
}
if (algorithmName.length <= MAX_ALGORITHMS) {
maxAPi = Double.MIN_VALUE;
minPi = Double.MAX_VALUE;
for (j = 0; j < exhaustiveI.size(); j++) {
if (exhaustiveI.elementAt(j).toString().contains(parejitas[order[i]].toString())) {
minPi =
2
* CDF_Normal.normp(
(-1)
* Math.abs(
Rj[
((Relation)
((Vector<Relation>) exhaustiveI.elementAt(j))
.elementAt(0))
.i]
- Rj[
((Relation)
((Vector<Relation>) exhaustiveI.elementAt(j))
.elementAt(0))
.j])
/ SE);
for (k = 1; k < ((Vector<Relation>) exhaustiveI.elementAt(j)).size(); k++) {
tmpPi =
2
* CDF_Normal.normp(
(-1)
* Math.abs(
Rj[
((Relation)
((Vector<Relation>) exhaustiveI.elementAt(j))
.elementAt(k))
.i]
- Rj[
((Relation)
((Vector<Relation>) exhaustiveI.elementAt(j))
.elementAt(k))
.j])
/ SE);
if (tmpPi < minPi) {
minPi = tmpPi;
}
}
tmpAPi = minPi * (double) (((Vector<Relation>) exhaustiveI.elementAt(j)).size());
if (tmpAPi > maxAPi) {
maxAPi = tmpAPi;
}
}
}
adjustedP[i][3] = maxAPi;
}
}
for (i = 1; i < adjustedP.length; i++) {
if (adjustedP[i][1] < adjustedP[i - 1][1]) adjustedP[i][1] = adjustedP[i - 1][1];
if (adjustedP[i][2] < adjustedP[i - 1][2]) adjustedP[i][2] = adjustedP[i - 1][2];
if (adjustedP[i][3] < adjustedP[i - 1][3]) adjustedP[i][3] = adjustedP[i - 1][3];
}
count = 3;
if (Nemenyi) {
count++;
}
if (Holm) {
count++;
}
if (Scha) {
count++;
}
if (Berg) {
count++;
}
out +=
"\\begin{table}[!htp]\n\\centering\\scriptsize\n"
+ "\\begin{tabular}{"
+ printC(count)
+ "}\n"
+ "i&hypothesis&unadjusted $p$";
if (Nemenyi) {
out += "&$p_{Neme}$";
}
if (Holm) {
out += "&$p_{Holm}$";
}
if (Scha) {
out += "&$p_{Shaf}$";
}
if (Berg) {
out += "&$p_{Berg}$";
}
out += "\\\\\n\\hline";
for (i = 0; i < Pi.length; i++) {
out +=
(i + 1)
+ "&"
+ algorithmName[parejitas[order[i]].i]
+ " vs ."
+ algorithmName[parejitas[order[i]].j]
+ "&"
+ nf6.format(Pi[i]);
if (Nemenyi) {
out += "&" + nf6.format(adjustedP[i][0]);
}
if (Holm) {
out += "&" + nf6.format(adjustedP[i][1]);
}
if (Scha) {
out += "&" + nf6.format(adjustedP[i][2]);
}
if (Berg) {
out += "&" + nf6.format(adjustedP[i][3]);
}
out += "\\\\\n";
}
out += "\\hline\n" + "\\end{tabular}\n\\caption{Adjusted $p$-values}\n" + "\\end{table}\n\n";
out += "\\end{landscape}\n\\end{document}";
return out;
} // end-method
/** Process the training and test files provided in the parameters file to the constructor. */
public void process() {
// declarations
double[] outputs;
double[] outputs2;
Instance neighbor;
double dist, mean;
int actual;
Randomize rnd = new Randomize();
Instance ex;
gCenter kmeans = null;
int iterations = 0;
double E;
double prevE;
int totalMissing = 0;
boolean allMissing = true;
rnd.setSeed(semilla);
// PROCESS
try {
// Load in memory a dataset that contains a classification problem
IS.readSet(input_train_name, true);
int in = 0;
int out = 0;
ndatos = IS.getNumInstances();
nvariables = Attributes.getNumAttributes();
nentradas = Attributes.getInputNumAttributes();
nsalidas = Attributes.getOutputNumAttributes();
X = new String[ndatos][nvariables]; // matrix with transformed data
kmeans = new gCenter(K, ndatos, nvariables);
timesSeen = new FreqList[nvariables];
mostCommon = new String[nvariables];
// first, we choose k 'means' randomly from all
// instances
totalMissing = 0;
for (int i = 0; i < ndatos; i++) {
Instance inst = IS.getInstance(i);
if (inst.existsAnyMissingValue()) totalMissing++;
}
if (totalMissing == ndatos) allMissing = true;
else allMissing = false;
for (int numMeans = 0; numMeans < K; numMeans++) {
do {
actual = (int) (ndatos * rnd.Rand());
ex = IS.getInstance(actual);
} while (ex.existsAnyMissingValue() && !allMissing);
kmeans.copyCenter(ex, numMeans);
}
// now, iterate adjusting clusters' centers and
// instances to them
prevE = 0;
iterations = 0;
do {
for (int i = 0; i < ndatos; i++) {
Instance inst = IS.getInstance(i);
kmeans.setClusterOf(inst, i);
}
// set new centers
kmeans.recalculateCenters(IS);
// compute RMSE
E = 0;
for (int i = 0; i < ndatos; i++) {
Instance inst = IS.getInstance(i);
E += kmeans.distance(inst, kmeans.getClusterOf(i));
}
iterations++;
// System.out.println(iterations+"\t"+E);
if (Math.abs(prevE - E) == 0) iterations = maxIter;
else prevE = E;
} while (E > minError && iterations < maxIter);
for (int i = 0; i < ndatos; i++) {
Instance inst = IS.getInstance(i);
in = 0;
out = 0;
for (int j = 0; j < nvariables; j++) {
Attribute a = Attributes.getAttribute(j);
direccion = a.getDirectionAttribute();
tipo = a.getType();
if (direccion == Attribute.INPUT) {
if (tipo != Attribute.NOMINAL && !inst.getInputMissingValues(in)) {
X[i][j] = new String(String.valueOf(inst.getInputRealValues(in)));
} else {
if (!inst.getInputMissingValues(in)) X[i][j] = inst.getInputNominalValues(in);
else {
actual = kmeans.getClusterOf(i);
X[i][j] = new String(kmeans.valueAt(actual, j));
}
}
in++;
} else {
if (direccion == Attribute.OUTPUT) {
if (tipo != Attribute.NOMINAL && !inst.getOutputMissingValues(out)) {
X[i][j] = new String(String.valueOf(inst.getOutputRealValues(out)));
} else {
if (!inst.getOutputMissingValues(out)) X[i][j] = inst.getOutputNominalValues(out);
else {
actual = kmeans.getClusterOf(i);
X[i][j] = new String(kmeans.valueAt(actual, j));
}
}
out++;
}
}
}
}
} catch (Exception e) {
System.out.println("Dataset exception = " + e);
e.printStackTrace();
System.exit(-1);
}
write_results(output_train_name);
/** ************************************************************************************ */
// does a test file associated exist?
if (input_train_name.compareTo(input_test_name) != 0) {
try {
// Load in memory a dataset that contains a classification problem
IStest.readSet(input_test_name, false);
int in = 0;
int out = 0;
ndatos = IStest.getNumInstances();
nvariables = Attributes.getNumAttributes();
nentradas = Attributes.getInputNumAttributes();
nsalidas = Attributes.getOutputNumAttributes();
for (int i = 0; i < ndatos; i++) {
Instance inst = IStest.getInstance(i);
in = 0;
out = 0;
for (int j = 0; j < nvariables; j++) {
Attribute a = Attributes.getAttribute(j);
direccion = a.getDirectionAttribute();
tipo = a.getType();
if (direccion == Attribute.INPUT) {
if (tipo != Attribute.NOMINAL && !inst.getInputMissingValues(in)) {
X[i][j] = new String(String.valueOf(inst.getInputRealValues(in)));
} else {
if (!inst.getInputMissingValues(in)) X[i][j] = inst.getInputNominalValues(in);
else {
actual = kmeans.getClusterOf(i);
X[i][j] = new String(kmeans.valueAt(actual, j));
}
}
in++;
} else {
if (direccion == Attribute.OUTPUT) {
if (tipo != Attribute.NOMINAL && !inst.getOutputMissingValues(out)) {
X[i][j] = new String(String.valueOf(inst.getOutputRealValues(out)));
} else {
if (!inst.getOutputMissingValues(out)) X[i][j] = inst.getOutputNominalValues(out);
else {
actual = kmeans.getClusterOf(i);
X[i][j] = new String(kmeans.valueAt(actual, j));
}
}
out++;
}
}
}
}
} catch (Exception e) {
System.out.println("Dataset exception = " + e);
e.printStackTrace();
System.exit(-1);
}
write_results(output_test_name);
}
}
/**
* Computes and stores the information gain of each attribute (variable) of the dataset
*
* @param Examples Set of instances of the dataset
* @param nFile Name of the file
*/
public void GainInit(TableDat Examples, String nFile) {
int i, j, h, v;
boolean encontrado;
float info_gk, suma, suma1, suma2, p_clase, logaritmo;
int num_clase[] = new int[n_clases];
int n_vars = this.getNVars();
int MaxVal = this.getMaxVal();
float p[][] = new float[n_vars][MaxVal];
float p_cond[][][] = new float[n_clases][n_vars][MaxVal];
GI = new float[n_vars];
intervalosGI = new float[n_vars][MaxVal];
String contents;
contents = "\n--------------------------------------------\n";
contents += "| Computation of the info gain |\n";
contents += "--------------------------------------------\n";
contents += "Points for computation of the info gain:\n";
// Loads the values for "intervalosGI"
float marca, p_corte;
for (int v1 = 0; v1 < n_vars; v1++) {
if (this.getContinuous(v1) == true) {
contents += "\tVariable " + var[v1].getName() + ": ";
marca = (this.getMax(v1) - this.getMin(v1)) / ((float) (this.getNLabelVar(v1) - 1));
p_corte = this.getMin(v1) + marca / 2;
for (int et = 0; et < this.getNLabelVar(v1); et++) {
intervalosGI[v1][et] = p_corte;
contents += intervalosGI[v1][et] + " ";
p_corte += marca;
}
contents += "\n";
}
}
// Structure initialization
for (i = 0; i < n_clases; i++) num_clase[i] = 0;
for (i = 0; i < n_vars; i++)
for (j = 0; j < MaxVal; j++) {
p[i][j] = 0; // Simple probabilities matrix
for (h = 0; h < n_clases; h++) p_cond[h][i][j] = 0; // Conditional probabilities matrix
}
// Computation of the Simple and Conditional probabilities matrixs
for (i = 0; i < Examples.getNEx(); i++) {
num_clase[Examples.getClass(i)]++; // distribution by classes
for (j = 0; j < n_vars; j++) { // distribution by values
if (!this.getContinuous(j)) { // Discrete variable
if (!Examples.getLost(this, i, j)) {
// if the value is not a lost one
p[j][(int) Examples.getDat(i, j)]++;
p_cond[(int) Examples.getClass(i)][j][(int) Examples.getDat(i, j)]++;
}
} else { // Continuous variable
encontrado = false;
h = 0;
while (!encontrado && h < this.getNLabelVar(j)) {
if (Examples.getDat(i, j) <= intervalosGI[j][h]) encontrado = true;
else h++;
}
if (encontrado == true) {
p[j][h]++;
p_cond[(int) Examples.getClass(i)][j][h]++;
} else {
if (!Examples.getLost(this, i, j)) {
// Lost value
System.out.println(
"Fallo al calcular la ganancia de infor, Variable " + j + " Ejemplo " + i);
return;
}
}
}
}
}
for (h = 0; h < n_clases; h++)
for (i = 0; i < n_vars; i++) {
if (!this.getContinuous(i)) // Discrete variable
for (j = (int) this.getMin(i); j <= (int) this.getMax(i); j++)
p_cond[h][i][j] = p_cond[h][i][j] / Examples.getNEx();
else // Continuous variable
for (j = 0; j < this.getNLabelVar(i); j++)
p_cond[h][i][j] = p_cond[h][i][j] / Examples.getNEx();
}
for (i = 0; i < n_vars; i++) {
if (!this.getContinuous(i)) // Discrete variable
for (j = (int) this.getMin(i); j <= (int) this.getMax(i); j++)
p[i][j] = p[i][j] / Examples.getNEx();
else // Continuous variable
for (j = 0; j < this.getNLabelVar(i); j++) p[i][j] = p[i][j] / Examples.getNEx();
}
// Info Gk computation
suma = 0;
for (i = 0; i < n_clases; i++) {
p_clase = ((float) num_clase[i]) / Examples.getNEx();
if (p_clase > 0) {
logaritmo = (float) (Math.log((double) p_clase) / Math.log(2));
suma += p_clase * logaritmo;
}
}
info_gk = (-1) * suma;
// Information gain computation for each attibute
for (v = 0; v < n_vars; v++) {
suma = info_gk;
suma1 = 0;
if (!this.getContinuous(v)) { // Discrete Variable
for (i = (int) this.getMin(v); i <= (int) this.getMax(v); i++) {
suma2 = 0;
for (j = 0; j < n_clases; j++)
if (p_cond[j][v][i] > 0) {
logaritmo = (float) (Math.log(p_cond[j][v][i]) / Math.log(2));
suma2 += p_cond[j][v][i] * logaritmo;
}
suma1 += p[v][i] * (-1) * suma2;
}
} else { // Continuous Variable
for (i = 0; i < this.getNLabelVar(v); i++) {
suma2 = 0;
for (j = 0; j < n_clases; j++)
if (p_cond[j][v][i] > 0) {
logaritmo = (float) (Math.log(p_cond[j][v][i]) / Math.log(2));
suma2 += p_cond[j][v][i] * logaritmo;
}
suma1 += p[v][i] * (-1) * suma2;
}
}
GI[v] = suma + (-1) * suma1;
}
contents += "Information Gain of the variables:\n";
for (v = 0; v < n_vars; v++) {
if (this.getContinuous(v) == true)
contents += "\tVariable " + var[v].getName() + ": " + GI[v] + "\n";
}
if (nFile != "") Files.addToFile(nFile, contents);
}
/** It launches the algorithm */
public void execute() {
int i, j, k, l;
int t;
int ele;
double prob[];
double aux;
double NUmax = 1.5; // used for lineal ranking
double NUmin = 0.5; // used for lineal ranking
double pos1, pos2;
int sel1, sel2;
int data[][];
int infoAttr[];
int classData[];
Vector<Rule> contenedor = new Vector<Rule>();
Vector<Rule> conjR = new Vector<Rule>();
Rule tmpRule;
Condition tmpCondition[] = new Condition[1];
RuleSet population[];
RuleSet hijo1, hijo2;
if (somethingWrong) { // We do not execute the program
System.err.println("An error was found, the data-set has numerical values.");
System.err.println("Aborting the program");
// We should not use the statement: System.exit(-1);
} else {
Randomize.setSeed(seed);
nClasses = train.getnClasses();
/*Build the nominal data information*/
infoAttr = new int[train.getnInputs()];
for (i = 0; i < infoAttr.length; i++) {
infoAttr[i] = train.numberValues(i);
}
data = new int[train.getnData()][train.getnInputs()];
for (i = 0; i < data.length; i++) {
for (j = 0; j < data[i].length; j++) {
if (train.isMissing(i, j)) data[i][j] = -1;
else data[i][j] = train.valueExample(i, j);
}
}
classData = new int[train.getnData()];
for (i = 0; i < classData.length; i++) {
classData[i] = train.getOutputAsInteger(i);
}
/*Find first-order rules which result interesting*/
for (i = 0; i < nClasses; i++) {
for (j = 0; j < infoAttr.length; j++) {
for (k = 0; k < infoAttr[j]; k++) {
tmpCondition[0] = new Condition(j, k);
tmpRule = new Rule(tmpCondition);
if (Math.abs(computeAdjustedResidual(data, classData, tmpRule, i)) > 1.96) {
if (!contenedor.contains(tmpRule)) {
contenedor.add(tmpRule);
conjR.add(tmpRule);
}
}
}
}
}
// Construct the Baker selection roulette
prob = new double[popSize];
for (j = 0; j < popSize; j++) {
aux = (double) (NUmax - NUmin) * ((double) j / (popSize - 1));
prob[j] = (double) (1.0 / (popSize)) * (NUmax - aux);
}
for (j = 1; j < popSize; j++) prob[j] = prob[j] + prob[j - 1];
/*Steady-State Genetic Algorithm*/
ele = 2;
population = new RuleSet[popSize];
while (conjR.size() >= 2) {
t = 0;
System.out.println("Producing rules of level " + ele);
for (i = 0; i < population.length; i++) {
population[i] = new RuleSet(conjR);
population[i].computeFitness(data, classData, infoAttr, contenedor, nClasses);
}
Arrays.sort(population);
while (t < numGenerations && !population[0].equals(population[popSize - 1])) {
System.out.println("Generation " + t);
t++;
/*Baker's selection*/
pos1 = Randomize.Rand();
pos2 = Randomize.Rand();
for (l = 0; l < popSize && prob[l] < pos1; l++) ;
sel1 = l;
for (l = 0; l < popSize && prob[l] < pos2; l++) ;
sel2 = l;
hijo1 = new RuleSet(population[sel1]);
hijo2 = new RuleSet(population[sel2]);
if (Randomize.Rand() < pCross) {
RuleSet.crossover1(hijo1, hijo2);
} else {
RuleSet.crossover2(hijo1, hijo2);
}
RuleSet.mutation(hijo1, conjR, pMut, data, classData, infoAttr, contenedor, nClasses);
RuleSet.mutation(hijo2, conjR, pMut, data, classData, infoAttr, contenedor, nClasses);
hijo1.computeFitness(data, classData, infoAttr, contenedor, nClasses);
hijo2.computeFitness(data, classData, infoAttr, contenedor, nClasses);
population[popSize - 2] = new RuleSet(hijo1);
population[popSize - 1] = new RuleSet(hijo2);
Arrays.sort(population);
}
/*Decode function*/
ele++;
conjR.removeAllElements();
System.out.println(
"Fitness of the best chromosome in rule level " + ele + ": " + population[0].fitness);
for (i = 0; i < population[0].getRuleSet().length; i++) {
if (Math.abs(computeAdjustedResidual(data, classData, population[0].getRule(i), i))
> 1.96) {
if (validarRegla(population[0].getRule(i))
&& !contenedor.contains(population[0].getRule(i))) {
contenedor.add(population[0].getRule(i));
conjR.add(population[0].getRule(i));
}
}
}
}
// Finally we should fill the training and test output files
doOutput(this.val, this.outputTr, data, classData, infoAttr, contenedor, nClasses);
doOutput(this.test, this.outputTst, data, classData, infoAttr, contenedor, nClasses);
/*Print the rule obtained*/
for (i = contenedor.size() - 1; i >= 0; i--) {
if (reglaPositiva(
this.train, data, classData, infoAttr, nClasses, contenedor.elementAt(i))) {
Fichero.AnadirtoFichero(outputRule, contenedor.elementAt(i).toString(train));
Fichero.AnadirtoFichero(
outputRule,
" -> "
+ consecuente(
this.train, data, classData, infoAttr, nClasses, contenedor.elementAt(i))
+ "\n");
}
}
System.out.println("Algorithm Finished");
}
}
/**
* SMOTE preprocessing procedure
*
* @param datosTrain input training dta
* @param realTrain actual training data
* @param nominalTrain nominal attribute values
* @param nulosTrain null values
* @param clasesTrain training classes
* @param datosArt synthetic instances
*/
public void SMOTE(
double datosTrain[][],
double realTrain[][],
int nominalTrain[][],
boolean nulosTrain[][],
int clasesTrain[],
double datosArt[][],
double realArt[][],
int nominalArt[][],
boolean nulosArt[][],
int clasesArt[],
int kSMOTE,
int ASMO,
double smoting,
boolean balance,
int nPos,
int posID,
int nNeg,
int negID,
boolean distanceEu) {
int i, j, l, m;
int tmp, pos;
int positives[];
int neighbors[][];
double genS[][];
double genR[][];
int genN[][];
boolean genM[][];
int clasesGen[];
int nn;
/* Localize the positive instances */
positives = new int[nPos];
for (i = 0, j = 0; i < clasesTrain.length; i++) {
if (clasesTrain[i] == posID) {
positives[j] = i;
j++;
}
}
/* Randomize the instance presentation */
for (i = 0; i < positives.length; i++) {
tmp = positives[i];
pos = Randomize.Randint(0, positives.length - 1);
positives[i] = positives[pos];
positives[pos] = tmp;
}
/* Obtain k-nearest neighbors of each positive instance */
neighbors = new int[positives.length][kSMOTE];
for (i = 0; i < positives.length; i++) {
switch (ASMO) {
case 0:
KNN.evaluacionKNN2(
kSMOTE,
datosTrain,
realTrain,
nominalTrain,
nulosTrain,
clasesTrain,
datosTrain[positives[i]],
realTrain[positives[i]],
nominalTrain[positives[i]],
nulosTrain[positives[i]],
Math.max(posID, negID) + 1,
distanceEu,
neighbors[i]);
break;
case 1:
evaluacionKNNClass(
kSMOTE,
datosTrain,
realTrain,
nominalTrain,
nulosTrain,
clasesTrain,
datosTrain[positives[i]],
realTrain[positives[i]],
nominalTrain[positives[i]],
nulosTrain[positives[i]],
Math.max(posID, negID) + 1,
distanceEu,
neighbors[i],
posID);
break;
case 2:
evaluacionKNNClass(
kSMOTE,
datosTrain,
realTrain,
nominalTrain,
nulosTrain,
clasesTrain,
datosTrain[positives[i]],
realTrain[positives[i]],
nominalTrain[positives[i]],
nulosTrain[positives[i]],
Math.max(posID, negID) + 1,
distanceEu,
neighbors[i],
negID);
break;
}
}
/* Interpolation of the minority instances */
if (balance) {
genS = new double[nNeg - nPos][datosTrain[0].length];
genR = new double[nNeg - nPos][datosTrain[0].length];
genN = new int[nNeg - nPos][datosTrain[0].length];
genM = new boolean[nNeg - nPos][datosTrain[0].length];
clasesGen = new int[nNeg - nPos];
} else {
genS = new double[(int) (nPos * smoting)][datosTrain[0].length];
genR = new double[(int) (nPos * smoting)][datosTrain[0].length];
genN = new int[(int) (nPos * smoting)][datosTrain[0].length];
genM = new boolean[(int) (nPos * smoting)][datosTrain[0].length];
clasesGen = new int[(int) (nPos * smoting)];
}
for (i = 0; i < genS.length; i++) {
clasesGen[i] = posID;
nn = Randomize.Randint(0, kSMOTE - 1);
interpola(
realTrain[positives[i % positives.length]],
realTrain[neighbors[i % positives.length][nn]],
nominalTrain[positives[i % positives.length]],
nominalTrain[neighbors[i % positives.length][nn]],
nulosTrain[positives[i % positives.length]],
nulosTrain[neighbors[i % positives.length][nn]],
genS[i],
genR[i],
genN[i],
genM[i]);
}
for (j = 0; j < datosTrain.length; j++) {
for (l = 0; l < datosTrain[0].length; l++) {
datosArt[j][l] = datosTrain[j][l];
realArt[j][l] = realTrain[j][l];
nominalArt[j][l] = nominalTrain[j][l];
nulosArt[j][l] = nulosTrain[j][l];
}
clasesArt[j] = clasesTrain[j];
}
for (m = 0; j < datosArt.length; j++, m++) {
for (l = 0; l < datosTrain[0].length; l++) {
datosArt[j][l] = genS[m][l];
realArt[j][l] = genR[m][l];
nominalArt[j][l] = genN[m][l];
nulosArt[j][l] = genM[m][l];
}
clasesArt[j] = clasesGen[m];
}
}
private double computeStandarizedResidual(
int data[][], int classData[], Condition cond, int clase) {
double tmp = computeEAipAjq(data, classData, cond, clase);
return (computeCountAipAjq(data, classData, cond, clase) - tmp) / Math.sqrt(tmp);
}
public void ejecutar() {
int i, j, l, m;
double alfai;
int nClases;
int claseObt;
boolean marcas[];
boolean notFound;
int init;
int clasSel[];
int baraje[];
int pos, tmp;
String instanciasIN[];
String instanciasOUT[];
long tiempo = System.currentTimeMillis();
/* Getting the number of differents classes */
nClases = 0;
for (i = 0; i < clasesTrain.length; i++) if (clasesTrain[i] > nClases) nClases = clasesTrain[i];
nClases++;
/* Shuffle the train set */
baraje = new int[datosTrain.length];
Randomize.setSeed(semilla);
for (i = 0; i < datosTrain.length; i++) baraje[i] = i;
for (i = 0; i < datosTrain.length; i++) {
pos = Randomize.Randint(i, datosTrain.length - 1);
tmp = baraje[i];
baraje[i] = baraje[pos];
baraje[pos] = tmp;
}
/*
* Inicialization of the flagged instaces vector for a posterior
* elimination
*/
marcas = new boolean[datosTrain.length];
for (i = 0; i < datosTrain.length; i++) marcas[i] = false;
if (datosTrain.length > 0) {
// marcas[baraje[0]] = true; //the first instance is included always
nSel = n_p;
if (nSel < nClases) nSel = nClases;
} else {
System.err.println("Input dataset is empty");
nSel = 0;
}
clasSel = new int[nClases];
System.out.print("Selecting initial neurons... ");
// at least, there must be 1 neuron of each class at the beginning
init = nClases;
for (i = 0; i < nClases && i < datosTrain.length; i++) {
pos = Randomize.Randint(0, datosTrain.length - 1);
tmp = 0;
while ((clasesTrain[pos] != i || marcas[pos]) && tmp < datosTrain.length) {
pos = (pos + 1) % datosTrain.length;
tmp++;
}
if (tmp < datosTrain.length) marcas[pos] = true;
else init--;
// clasSel[i] = i;
}
for (i = init; i < Math.min(nSel, datosTrain.length); i++) {
tmp = 0;
pos = Randomize.Randint(0, datosTrain.length - 1);
while (marcas[pos]) {
pos = (pos + 1) % datosTrain.length;
tmp++;
}
// if(i<nClases){
// notFound = true;
// do{
// for(j=i-1;j>=0 && notFound;j--){
// if(clasSel[j] == clasesTrain[pos])
// notFound = false;
// }
// if(!notFound)
// pos = Randomize.Randint (0, datosTrain.length-1);
// }while(!notFound);
// }
// clasSel[i] = clasesTrain[pos];
marcas[pos] = true;
init++;
}
nSel = init;
System.out.println("Initial neurons selected: " + nSel);
/* Building of the S set from the flags */
conjS = new double[nSel][datosTrain[0].length];
clasesS = new int[nSel];
for (m = 0, l = 0; m < datosTrain.length; m++) {
if (marcas[m]) { // the instance must be copied to the solution
for (j = 0; j < datosTrain[0].length; j++) {
conjS[l][j] = datosTrain[m][j];
}
clasesS[l] = clasesTrain[m];
l++;
}
}
alfai = alpha;
boolean change = true;
/* Body of the LVQ algorithm. */
// Train the network
for (int it = 0; it < T && change; it++) {
change = false;
alpha = alfai;
for (i = 1; i < datosTrain.length; i++) {
// search for the nearest neuron to training instance
pos = NN(nSel, conjS, datosTrain[baraje[i]]);
// nearest neuron labels correctly the class of training
// instance?
if (clasesS[pos] != clasesTrain[baraje[i]]) { // NO - repel
// the neuron
for (j = 0; j < conjS[pos].length; j++) {
conjS[pos][j] = conjS[pos][j] - alpha * (datosTrain[baraje[i]][j] - conjS[pos][j]);
}
change = true;
} else { // YES - migrate the neuron towards the input vector
for (j = 0; j < conjS[pos].length; j++) {
conjS[pos][j] = conjS[pos][j] + alpha * (datosTrain[baraje[i]][j] - conjS[pos][j]);
}
}
alpha = nu * alpha;
}
// Shuffle again the training partition
baraje = new int[datosTrain.length];
for (i = 0; i < datosTrain.length; i++) baraje[i] = i;
for (i = 0; i < datosTrain.length; i++) {
pos = Randomize.Randint(i, datosTrain.length - 1);
tmp = baraje[i];
baraje[i] = baraje[pos];
baraje[pos] = tmp;
}
}
System.out.println(
"LVQ " + relation + " " + (double) (System.currentTimeMillis() - tiempo) / 1000.0 + "s");
// Classify the train data set
instanciasIN = new String[datosReferencia.length];
instanciasOUT = new String[datosReferencia.length];
for (i = 0; i < datosReferencia.length; i++) {
/* Classify the instance selected in this iteration */
Attribute a = Attributes.getOutputAttribute(0);
int tipo = a.getType();
claseObt = KNN.evaluacionKNN2(1, conjS, clasesS, datosReferencia[i], nClases);
if (tipo != Attribute.NOMINAL) {
instanciasIN[i] = new String(String.valueOf(clasesReferencia[i]));
instanciasOUT[i] = new String(String.valueOf(claseObt));
} else {
instanciasIN[i] = new String(a.getNominalValue(clasesReferencia[i]));
instanciasOUT[i] = new String(a.getNominalValue(claseObt));
}
}
escribeSalida(
ficheroSalida[0], instanciasIN, instanciasOUT, entradas, salida, nEntradas, relation);
// Classify the test data set
normalizarTest();
instanciasIN = new String[datosTest.length];
instanciasOUT = new String[datosTest.length];
for (i = 0; i < datosTest.length; i++) {
/* Classify the instance selected in this iteration */
Attribute a = Attributes.getOutputAttribute(0);
int tipo = a.getType();
claseObt = KNN.evaluacionKNN2(1, conjS, clasesS, datosTest[i], nClases);
if (tipo != Attribute.NOMINAL) {
instanciasIN[i] = new String(String.valueOf(clasesTest[i]));
instanciasOUT[i] = new String(String.valueOf(claseObt));
} else {
instanciasIN[i] = new String(a.getNominalValue(clasesTest[i]));
instanciasOUT[i] = new String(a.getNominalValue(claseObt));
}
}
escribeSalida(
ficheroSalida[1], instanciasIN, instanciasOUT, entradas, salida, nEntradas, relation);
// Print the network to a file
printNetworkToFile(ficheroSalida[2], referencia.getHeader());
}
/**
* Obtain all exhaustive comparisons possible from an array of indexes
*
* @param indices A verctos of indexes.
* @return A vector with vectors containing all the possible relations between the indexes
*/
@SuppressWarnings("unchecked")
public static Vector<Vector<Relation>> obtainExhaustive(Vector<Integer> indices) {
Vector<Vector<Relation>> result = new Vector<Vector<Relation>>();
int i, j, k;
String binario;
boolean[] number = new boolean[indices.size()];
Vector<Integer> ind1, ind2;
Vector<Relation> set = new Vector<Relation>();
Vector<Vector<Relation>> res1, res2;
Vector<Relation> temp;
Vector<Relation> temp2;
Vector<Relation> temp3;
ind1 = new Vector<Integer>();
ind2 = new Vector<Integer>();
temp = new Vector<Relation>();
temp2 = new Vector<Relation>();
temp3 = new Vector<Relation>();
for (i = 0; i < indices.size(); i++) {
for (j = i + 1; j < indices.size(); j++) {
set.addElement(
new Relation(
((Integer) indices.elementAt(i)).intValue(),
((Integer) indices.elementAt(j)).intValue()));
}
}
if (set.size() > 0) result.addElement(set);
for (i = 1; i < (int) (Math.pow(2, indices.size() - 1)); i++) {
Arrays.fill(number, false);
ind1.removeAllElements();
ind2.removeAllElements();
temp.removeAllElements();
temp2.removeAllElements();
temp3.removeAllElements();
binario = Integer.toString(i, 2);
for (k = 0; k < number.length - binario.length(); k++) {
number[k] = false;
}
for (j = 0; j < binario.length(); j++, k++) {
if (binario.charAt(j) == '1') number[k] = true;
}
for (j = 0; j < number.length; j++) {
if (number[j] == true) {
ind1.addElement(new Integer(((Integer) indices.elementAt(j)).intValue()));
} else {
ind2.addElement(new Integer(((Integer) indices.elementAt(j)).intValue()));
}
}
res1 = obtainExhaustive(ind1);
res2 = obtainExhaustive(ind2);
for (j = 0; j < res1.size(); j++) {
result.addElement(new Vector<Relation>((Vector<Relation>) res1.elementAt(j)));
}
for (j = 0; j < res2.size(); j++) {
result.addElement(new Vector<Relation>((Vector<Relation>) res2.elementAt(j)));
}
for (j = 0; j < res1.size(); j++) {
temp = (Vector<Relation>) ((Vector<Relation>) res1.elementAt(j)).clone();
for (k = 0; k < res2.size(); k++) {
temp2 = (Vector<Relation>) temp.clone();
temp3 = (Vector<Relation>) ((Vector<Relation>) res2.elementAt(k)).clone();
if (((Relation) temp2.elementAt(0)).i < ((Relation) temp3.elementAt(0)).i) {
temp2.addAll((Vector<Relation>) temp3);
result.addElement(new Vector<Relation>(temp2));
} else {
temp3.addAll((Vector<Relation>) temp2);
result.addElement(new Vector<Relation>(temp3));
}
}
}
}
for (i = 0; i < result.size(); i++) {
if (((Vector<Relation>) result.elementAt(i)).toString().equalsIgnoreCase("[]")) {
result.removeElementAt(i);
i--;
}
}
for (i = 0; i < result.size(); i++) {
for (j = i + 1; j < result.size(); j++) {
if (((Vector<Relation>) result.elementAt(i))
.toString()
.equalsIgnoreCase(((Vector<Relation>) result.elementAt(j)).toString())) {
result.removeElementAt(j);
j--;
}
}
}
return result;
} // end-method
