public double getNearestInRangeDOFVal(
double startVal, double min, double max, DoubleMatrix1D x, int dof, int[] indices) {
// Return the nearest value of DOF #dof in x to startVal that will make this GenCoord (operating
// on
// the given indices of x) return a value in the range[min,max]
// If there is no such value, NaN is returned
switch (type) {
case REGULAR:
if (startVal < min) return min;
else if (startVal > max) return max;
else // startVal is in range already
return startVal;
case SUMSQ:
double sum = 0;
for (int q : indices) sum += Math.pow(x.get(q), 2);
double minsq = Math.pow(min, 2);
if (sum
< minsq) { // Increase the absolute value of x.get(dof) until it makes sum equal to
// min^2
double absAns = Math.sqrt(Math.pow(min, 2) - (sum - Math.pow(startVal, 2)));
if (startVal > 0) return absAns;
else return -absAns;
}
double maxsq = Math.pow(max, 2);
if (sum
> maxsq) { // Decrease the absolute value of x.get(dof) until it makes sum equal to
// min^2
double absAns = Math.sqrt(Math.pow(max, 2) - (sum - Math.pow(startVal, 2)));
if (startVal > 0) return absAns;
else return -absAns;
} else // In range already
return startVal;
case LINCOMB:
double ans = 0;
int DOFLocalInd = -1; // Index of dof in the "indices" array
for (int a = 0; a < coeffs.length; a++) {
if (indices[a] == dof) {
ans += coeffs[a] * startVal;
DOFLocalInd = a;
} else ans += coeffs[a] * x.get(indices[a]);
}
if (ans < min) {
if (coeffs[DOFLocalInd] == 0) return Double.NaN;
return startVal + (min - ans) / coeffs[DOFLocalInd];
} else if (ans > max) {
if (coeffs[DOFLocalInd] == 0) return Double.NaN;
return startVal + (max - ans) / coeffs[DOFLocalInd];
} else // In range already
return startVal;
default:
System.err.println("ERROR: Unrecognized generalized coordinate type: " + type);
System.exit(1);
return 0;
}
}
private double multiLL(DoubleMatrix2D coeffs, Node dep, List<Node> indep) {
DoubleMatrix2D indepData =
factory2D.make(internalData.subsetColumns(indep).getDoubleData().toArray());
List<Node> depList = new ArrayList<>();
depList.add(dep);
DoubleMatrix2D depData =
factory2D.make(internalData.subsetColumns(depList).getDoubleData().toArray());
int N = indepData.rows();
DoubleMatrix2D probs =
Algebra.DEFAULT.mult(factory2D.appendColumns(factory2D.make(N, 1, 1.0), indepData), coeffs);
probs =
factory2D
.appendColumns(factory2D.make(indepData.rows(), 1, 1.0), probs)
.assign(Functions.exp);
double ll = 0;
for (int i = 0; i < N; i++) {
DoubleMatrix1D curRow = probs.viewRow(i);
curRow.assign(Functions.div(curRow.zSum()));
ll += Math.log(curRow.get((int) depData.get(i, 0)));
}
return ll;
}
public double eval(DoubleMatrix1D x, int[] indices) {
// The "regular" coordinates to be used have the indicated indices in x
switch (type) {
case REGULAR:
return x.get(indices[0]);
case SUMSQ:
double sum = 0;
for (int q : indices) sum += Math.pow(x.get(q), 2);
return Math.sqrt(sum);
case LINCOMB:
double ans = 0;
for (int a = 0; a < coeffs.length; a++) ans += coeffs[a] * x.get(indices[a]);
return ans;
default:
System.err.println("ERROR: Unrecognized generalized coordinate type: " + type);
System.exit(1);
return 0;
}
}
/** @see "Convex Optimization, p. 610" */
protected DoubleMatrix1D rDual(DoubleMatrix1D gradF0X, DoubleMatrix1D L, DoubleMatrix1D V) {
// m1 = GradFiX[T].L + gradF0X
DoubleMatrix1D m1 = F1.make(getDim());
for (int i = 0; i < getDim(); i++) {
double m = 0;
m += -L.getQuick(i);
m += L.getQuick(getDim() + i);
m1.setQuick(i, m + gradF0X.get(i));
}
if (getMeq() == 0) {
return m1;
}
return ColtUtils.zMultTranspose(getA(), V, m1, 1.);
}
public static DoubleMatrix1D normalizeVector(DoubleMatrix1D vector) {
ArrayList<Double> vl = new ArrayList<Double>();
for (int i = 0; i < vector.size(); i++) {
vl.add(Math.pow(vector.get(i), 2));
}
vl = normalizeVector(vl);
DoubleMatrix1D rt = DoubleFactory1D.sparse.make(vl.size());
for (int i = 0; i < vector.size(); i++) {
rt.set(i, vl.get(i));
}
return rt;
}
public static boolean isInScope(
final DoubleMatrix1D position,
final DoubleMatrix1D dimensions,
final DoubleMatrix1D feature) {
if (position.size() != dimensions.size() || position.size() != feature.size()) {
throw new IllegalArgumentException(
"feature, position and dimensions" + "must have the same size");
}
// is the feature inside the tile (including boundaries)
for (int i = 0; i < feature.size(); i++) {
if (!(feature.get(i) >= position.get(i) - dimensions.get(i) / 2
&& feature.get(i) <= position.get(i) + dimensions.get(i) / 2)) {
return false;
}
}
return true;
}
public void decrementDisp(double x, double y) {
disp.set(0, disp.get(0) - x);
disp.set(1, disp.get(1) - y);
}
public void incrementDisp(double x, double y) {
disp.set(0, disp.get(0) + x);
disp.set(1, disp.get(1) + y);
}
public double getYDisp() {
return disp.get(1);
}
public double getXDisp() {
return disp.get(0);
}
/**
* Returns the best cut of a graph w.r.t. the degree of dissimilarity between points of different
* partitions and the degree of similarity between points of the same partition.
*
* @param W the weight matrix of the graph
* @return an array of two elements, each of these contains the points of a partition
*/
protected static int[][] bestCut(DoubleMatrix2D W) {
int n = W.columns();
// Builds the diagonal matrices D and D^(-1/2) (represented as their diagonals)
DoubleMatrix1D d = DoubleFactory1D.dense.make(n);
DoubleMatrix1D d_minus_1_2 = DoubleFactory1D.dense.make(n);
for (int i = 0; i < n; i++) {
double d_i = W.viewRow(i).zSum();
d.set(i, d_i);
d_minus_1_2.set(i, 1 / Math.sqrt(d_i));
}
DoubleMatrix2D D = DoubleFactory2D.sparse.diagonal(d);
// System.out.println("DoubleMatrix2D :\n"+D.toString());
DoubleMatrix2D X = D.copy();
// System.out.println("DoubleMatrix2D copy :\n"+X.toString());
// X = D^(-1/2) * (D - W) * D^(-1/2)
X.assign(W, Functions.minus);
// System.out.println("DoubleMatrix2D X: (D-W) :\n"+X.toString());
for (int i = 0; i < n; i++)
for (int j = 0; j < n; j++)
X.set(i, j, X.get(i, j) * d_minus_1_2.get(i) * d_minus_1_2.get(j));
// Computes the eigenvalues and the eigenvectors of X
EigenvalueDecomposition e = new EigenvalueDecomposition(X);
DoubleMatrix1D lambda = e.getRealEigenvalues();
// Selects the eigenvector z_2 associated with the second smallest eigenvalue
// Creates a map that contains the pairs <index, eigenvalue>
AbstractIntDoubleMap map = new OpenIntDoubleHashMap(n);
for (int i = 0; i < n; i++) map.put(i, Math.abs(lambda.get(i)));
IntArrayList list = new IntArrayList();
// Sorts the map on the value
map.keysSortedByValue(list);
// Gets the index of the second smallest element
int i_2 = list.get(1);
// y_2 = D^(-1/2) * z_2
DoubleMatrix1D y_2 = e.getV().viewColumn(i_2).copy();
y_2.assign(d_minus_1_2, Functions.mult);
// Creates a map that contains the pairs <i, y_2[i]>
map.clear();
for (int i = 0; i < n; i++) map.put(i, y_2.get(i));
// Sorts the map on the value
map.keysSortedByValue(list);
// Search the element in the map previuosly ordered that minimizes the cut
// of the partition
double best_cut = Double.POSITIVE_INFINITY;
int[][] partition = new int[2][];
// The array v contains all the elements of the graph ordered by their
// projection on vector y_2
int[] v = list.elements();
// For each admissible splitting point i
for (int i = 1; i < n; i++) {
// The array a contains all the elements that have a projection on vector
// y_2 less or equal to the one of i-th element
// The array b contains the remaining elements
int[] a = new int[i];
int[] b = new int[n - i];
System.arraycopy(v, 0, a, 0, i);
System.arraycopy(v, i, b, 0, n - i);
double cut = Ncut(W, a, b, v);
if (cut < best_cut) {
best_cut = cut;
partition[0] = a;
partition[1] = b;
}
}
// System.out.println("Partition:");
// UtilsJS.printMatrix(partition);
return partition;
}
/**
* Solves a presolved standard form LP problem in the form of min(c) s.t. A.x = b lb <= x <= ub
*/
protected int optimizePresolvedStandardLP() throws Exception {
log.info("optimizePresolvedStandardLP");
long tStart = System.currentTimeMillis();
LPOptimizationRequest lpRequest = getLPOptimizationRequest();
if (log.isDebugEnabled() && lpRequest.isDumpProblem()) {
log.debug("LP problem: " + lpRequest.toString());
}
if (this.dim <= -1) {
if (getLb().size() != getUb().size()) {
log.error("Lower and upper bounds must have the same dimension");
throw new IllegalArgumentException("Lower and upper bounds must have the same dimension");
}
this.dim = getLb().size();
double minDeltaBoundsValue = Double.MAX_VALUE;
int minDeltaBoundsIndex = -1;
for (int i = 0; i < getDim(); i++) {
double deltai = getUb().getQuick(i) - getLb().getQuick(i);
if (deltai < minDeltaBoundsValue) {
minDeltaBoundsValue = deltai;
minDeltaBoundsIndex = i;
}
}
log.info("min delta bounds index: " + minDeltaBoundsIndex);
log.info("min delta bounds value: " + minDeltaBoundsValue);
}
// this.boundedLb = new boolean[getDim()];
// this.boundedUb = new boolean[getDim()];
// for(int i=0; i<getDim(); i++){
// if(!isLbUnbounded(getLb().getQuick(i))){
// boundedLb[i] = true;
// nOfBoundedLb++;
// }
// if(!isUbUnbounded(getUb().getQuick(i))){
// boundedUb[i] = true;
// nOfBoundedUb++;
// }
// }
this.meq = (this.meq > -1) ? this.meq : ((getA() != null) ? getA().rows() : 0);
// this.mieq = (this.mieq>-1)? this.mieq : (nOfBoundedLb+nOfBoundedUb);
this.mieq = (this.mieq > -1) ? this.mieq : (2 * getDim());
if (log.isDebugEnabled()) {
log.debug("dim : " + getDim());
log.debug("meq : " + getMeq());
log.debug("mieq: " + getMieq());
}
LPOptimizationResponse lpResponse = new LPOptimizationResponse();
DoubleMatrix1D X0 = getInitialPoint();
if (X0 == null) {
DoubleMatrix1D X0NF = getNotFeasibleInitialPoint();
if (X0NF != null) {
double rPriX0NFNorm = Math.sqrt(ALG.norm2(rPri(X0NF)));
DoubleMatrix1D fiX0NF = getFi(X0NF);
int maxIndex = Utils.getMaxIndex(fiX0NF);
double maxValue = fiX0NF.get(maxIndex);
if (log.isDebugEnabled()) {
log.debug("rPriX0NFNorm : " + rPriX0NFNorm);
log.debug("X0NF : " + ArrayUtils.toString(X0NF.toArray()));
log.debug("fiX0NF : " + ArrayUtils.toString(fiX0NF.toArray()));
}
if (maxValue < 0 && rPriX0NFNorm <= getToleranceFeas()) {
// the provided not-feasible starting point is already feasible
log.debug("the provided initial point is already feasible");
X0 = X0NF;
}
}
if (X0 == null) {
BasicPhaseILPPDM bf1 = new BasicPhaseILPPDM(this);
X0 = bf1.findFeasibleInitialPoint();
}
}
// check X0 feasibility
DoubleMatrix1D fiX0 = getFi(X0);
int maxIndex = Utils.getMaxIndex(fiX0);
double maxValue = fiX0.get(maxIndex);
double rPriX0Norm = Math.sqrt(ALG.norm2(rPri(X0)));
if (maxValue >= 0. || rPriX0Norm > getToleranceFeas()) { // must be fi STRICTLY < 0
log.warn("rPriX0Norm : " + rPriX0Norm);
log.warn("ineqX0 : " + ArrayUtils.toString(fiX0.toArray()));
log.warn("max ineq index: " + maxIndex);
log.warn("max ineq value: " + maxValue);
// the point must be INTERNAL, fi are used as denominators
throw new Exception("initial point must be strictly feasible");
}
DoubleMatrix1D V0 = F1.make(getMeq());
if (getYlb() != null && getYub() != null) {
// NB: the Lagrangian multipliers for eq. constraints used in this interior point method (v)
// are the opposite of the Lagrangian multipliers for eq. constraints used in the presolver
// (y)
// and so Ylb<=y<=Yub becomes -Yub<=v<=-Ylb
for (int i = 0; i < getMeq(); i++) {
double v0i = 0;
if (!isLbUnbounded(getYlb().getQuick(i))) {
if (!isUbUnbounded(getYub().getQuick(i))) {
v0i = -(getYub().getQuick(i) + getYlb().getQuick(i)) / 2;
} else {
v0i = -getYlb().getQuick(i);
}
} else {
if (!isUbUnbounded(getYub().getQuick(i))) {
v0i = -getYub().getQuick(i);
} else {
v0i = 0;
}
}
V0.setQuick(i, v0i);
}
}
DoubleMatrix1D L0 = getInitialLagrangian();
if (L0 != null) {
for (int j = 0; j < L0.size(); j++) {
// must be >0
if (L0.get(j) <= 0) {
throw new IllegalArgumentException("initial lagrangian must be strictly > 0");
}
}
} else {
L0 = F1.make(getMieq(), 1.); // must be >0 strictly
if (getZlb() != null && getZub() != null) {
// Zlb<= L <=Zub, meaning that:
// zlb[i] and zub[i] are the bounds on the Lagrangian of the constraint associated with
// lb[i]<x[i]<ub[i]
// note that zlb.size = zub.size = lb.size = ub.size (and = n of variables of the problem (=
// getDim())
// and that L.size = nOfBoundedLb + nOfBoundedUb (and in general < 2*getDim())
int cntLB = 0;
int cntUB = 0;
for (int i = 0; i < getDim(); i++) {
double zlbi =
(isLbUnbounded(getZlb().getQuick(i))) ? 0 : getZlb().getQuick(i); // L must be > 0
double zubi = (isUbUnbounded(getZub().getQuick(i))) ? 1 : getZub().getQuick(i);
L0.setQuick(cntLB, (zubi - zlbi) / 2);
cntLB++;
L0.setQuick(getDim() + cntUB, (zubi - zlbi) / 2);
cntUB++;
}
} else {
// inequalities comes from the pairs lower bounds-upper bounds
// in the calculation of the H matrix fro the KKT system, each pairs gives terms of the
// form:
// t = tl + tu
// tl = -L[i] / fi[i] for the lower bound
// tu = L[dim+i] / fi[dim+i] for the upper bound
// we want t = 1, and hence
// L[i] > -cc * fi[i]
// L[dim+i] = (1 + L[i] / fi[i]) * fi[dim+i]
// double cc = 10;
// int nOfLB = getMieq()/2;
// for (int i = 0; i < nOfLB; i++) {
// L0.setQuick(i, -cc * fiX0.getQuick(i));
// L0.setQuick(nOfLB + i, (1 - 10) * fiX0.getQuick(nOfLB + i));
// double sum = -L0.getQuick(i)/fiX0.getQuick(i)+L0.getQuick(nOfLB +
// i)/fiX0.getQuick(nOfLB + i);
// log.debug("sum["+i+"]: " + sum);
// }
}
}
if (log.isDebugEnabled()) {
log.debug("X0: " + ArrayUtils.toString(X0.toArray()));
log.debug("V0: " + ArrayUtils.toString(V0.toArray()));
log.debug("L0: " + ArrayUtils.toString(L0.toArray()));
}
if (log.isInfoEnabled()) {
log.info("toleranceFeas: " + getToleranceFeas());
log.info("tolerance : " + getTolerance());
}
DoubleMatrix1D X = X0;
DoubleMatrix1D V = V0;
DoubleMatrix1D L = L0;
double previousF0X = Double.NaN;
double previousRPriXNorm = Double.NaN;
double previousRDualXLVNorm = Double.NaN;
double previousSurrDG = Double.NaN;
double t;
int iteration = 0;
// List<DoubleMatrix1D> XList = new ArrayList<DoubleMatrix1D>();
// List<double[]> SList = new ArrayList<double[]>();
while (true) {
iteration++;
// iteration limit condition
if (iteration == getMaxIteration() + 1) {
lpResponse.setReturnCode(OptimizationResponse.FAILED);
log.error("Max iterations limit reached");
throw new Exception("Max iterations limit reached");
}
// XList.add(XList.size(), X);
double F0X = getF0(X);
if (log.isInfoEnabled()) {
log.info("iteration: " + iteration);
log.info("f0(X)=" + F0X);
}
if (log.isDebugEnabled()) {
log.debug("X=" + ArrayUtils.toString(X.toArray()));
log.debug("L=" + ArrayUtils.toString(L.toArray()));
log.debug("V=" + ArrayUtils.toString(V.toArray()));
}
// determine functions evaluations
DoubleMatrix1D gradF0X = getGradF0(X);
DoubleMatrix1D fiX = getFi(X);
log.debug("fiX=" + ArrayUtils.toString(fiX.toArray()));
// DoubleMatrix2D GradFiX = getGradFi(X);
// DoubleMatrix2D GradFiXOLD = getGradFiOLD(X);
// determine t
double surrDG = getSurrogateDualityGap(fiX, L);
t = getMu() * getMieq() / surrDG;
log.debug("t: " + t);
// determine residuals
DoubleMatrix1D rPriX = rPri(X);
DoubleMatrix1D rCentXLt = rCent(fiX, L, t);
DoubleMatrix1D rDualXLV = rDual(gradF0X, L, V);
// DoubleMatrix1D rDualXLVOLD = rDualOLD(GradFiXOLD, gradF0X, L, V);
// log.debug("delta: " + ALG.normInfinity(rDualXLVOLD.assign(rDualXLV, Functions.minus)));
double rPriXNorm = Math.sqrt(ALG.norm2(rPriX));
double rCentXLtNorm = Math.sqrt(ALG.norm2(rCentXLt));
double rDualXLVNorm = Math.sqrt(ALG.norm2(rDualXLV));
double normRXLVt =
Math.sqrt(Math.pow(rPriXNorm, 2) + Math.pow(rCentXLtNorm, 2) + Math.pow(rDualXLVNorm, 2));
// @TODO: set log.debug not log.info
log.info("rPri norm: " + rPriXNorm);
log.info("rCent norm: " + rCentXLtNorm);
log.info("rDual norm: " + rDualXLVNorm);
log.info("surrDG : " + surrDG);
// custom exit condition
if (checkCustomExitConditions(X)) {
lpResponse.setReturnCode(OptimizationResponse.SUCCESS);
break;
}
// exit condition
if (rPriXNorm <= getToleranceFeas()
&& rDualXLVNorm <= getToleranceFeas()
&& surrDG <= getTolerance()) {
lpResponse.setReturnCode(OptimizationResponse.SUCCESS);
break;
}
// progress conditions
if (isCheckProgressConditions()) {
if (!Double.isNaN(previousRPriXNorm)
&& !Double.isNaN(previousRDualXLVNorm)
&& !Double.isNaN(previousSurrDG)) {
if ((previousRPriXNorm <= rPriXNorm && rPriXNorm >= getToleranceFeas())
|| (previousRDualXLVNorm <= rDualXLVNorm && rDualXLVNorm >= getToleranceFeas())) {
log.error("No progress achieved, exit iterations loop without desired accuracy");
lpResponse.setReturnCode(OptimizationResponse.FAILED);
throw new Exception(
"No progress achieved, exit iterations loop without desired accuracy");
}
}
previousRPriXNorm = rPriXNorm;
previousRDualXLVNorm = rDualXLVNorm;
previousSurrDG = surrDG;
}
// compute primal-dual search direction
// a) prepare 11.55 system
DoubleMatrix2D Hpd = GradLSum(L, fiX);
// DoubleMatrix2D HpdOLD = GradLSumOLD(GradFiXOLD, L, fiX);
// log.debug("delta: " + ALG.normInfinity(HpdOLD.assign(Hpd, Functions.minus)));
DoubleMatrix1D gradSum = gradSum(t, fiX);
DoubleMatrix1D g = null;
// if(getAT()==null){
if (getA() == null) {
g = ColtUtils.add(gradF0X, gradSum);
} else {
// g = ColtUtils.add(ColtUtils.add(gradF0X, gradSum), ALG.mult(getAT(), V));
g =
ColtUtils.add(
ColtUtils.add(gradF0X, gradSum),
ColtUtils.zMultTranspose(getA(), V, F1.make(getDim()), 0));
}
// b) solving 11.55 system
if (this.kktSolver == null) {
this.kktSolver = new UpperDiagonalHKKTSolver(getDim(), lpRequest.isRescalingDisabled());
// this.kktSolver = new DiagonalHKKTSolver(getDim(), lpRequest.isRescalingDisabled());
}
if (isCheckKKTSolutionAccuracy()) {
kktSolver.setCheckKKTSolutionAccuracy(true);
kktSolver.setToleranceKKT(getToleranceKKT());
}
kktSolver.setHMatrix(Hpd);
kktSolver.setGVector(g);
if (getA() != null) {
kktSolver.setAMatrix(getA());
// kktSolver.setATMatrix(getAT());
kktSolver.setHVector(rPriX);
// if(rPriXNorm > getToleranceFeas()){
// kktSolver.setHVector(rPriX);
// }
}
DoubleMatrix1D[] sol = kktSolver.solve();
DoubleMatrix1D stepX = sol[0];
// double[] signa = new double[stepX.size()];
// for(int p=0; p<stepX.size(); p++){
// signa[p] = Math.signum(stepX.getQuick(p));
// }
// SList.add(SList.size(), signa);
DoubleMatrix1D stepV = (sol[1] != null) ? sol[1] : F1.make(0);
if (log.isDebugEnabled()) {
log.debug("stepX: " + ArrayUtils.toString(stepX.toArray()));
log.debug("stepV: " + ArrayUtils.toString(stepV.toArray()));
}
// c) solving for L
DoubleMatrix1D stepL = F1.make(getMieq());
DoubleMatrix1D gradFiStepX = gradFiStepX(stepX);
for (int i = 0; i < getMieq(); i++) {
stepL.setQuick(
i, (-L.getQuick(i) * gradFiStepX.getQuick(i) + rCentXLt.getQuick(i)) / fiX.getQuick(i));
}
if (log.isDebugEnabled()) {
log.debug("stepL: " + ArrayUtils.toString(stepL.toArray()));
}
// line search and update
// a) sMax computation
double sMax = Double.MAX_VALUE;
for (int j = 0; j < getMieq(); j++) {
if (stepL.get(j) < 0) {
sMax = Math.min(-L.get(j) / stepL.get(j), sMax);
}
}
sMax = Math.min(1, sMax);
double s = 0.99 * sMax;
// b) backtracking with f
DoubleMatrix1D X1 = F1.make(X.size());
DoubleMatrix1D L1 = F1.make(L.size());
DoubleMatrix1D V1 = F1.make(V.size());
DoubleMatrix1D fiX1 = null;
DoubleMatrix1D gradF0X1 = null;
// DoubleMatrix2D GradFiX1 = null;
// DoubleMatrix2D GradFiX1 = null;
DoubleMatrix1D rPriX1 = null;
DoubleMatrix1D rCentX1L1t = null;
DoubleMatrix1D rDualX1L1V1 = null;
int cnt = 0;
boolean areAllNegative = true;
while (cnt < 500) {
cnt++;
// X1 = X + s*stepX
X1 = stepX.copy().assign(Mult.mult(s)).assign(X, Functions.plus);
DoubleMatrix1D ineqValueX1 = getFi(X1);
areAllNegative = true;
for (int j = 0; areAllNegative && j < getMieq(); j++) {
areAllNegative = (Double.compare(ineqValueX1.get(j), 0.) < 0);
}
if (areAllNegative) {
break;
}
s = getBeta() * s;
}
if (!areAllNegative) {
// exited from the feasible region
throw new Exception("Optimization failed: impossible to remain within the faesible region");
}
log.debug("s: " + s);
// c) backtracking with norm
double previousNormRX1L1V1t = Double.NaN;
cnt = 0;
while (cnt < 500) {
cnt++;
X1 = ColtUtils.add(X, stepX, s);
L1 = ColtUtils.add(L, stepL, s);
V1 = ColtUtils.add(V, stepV, s);
if (isInDomainF0(X1)) {
fiX1 = getFi(X1);
gradF0X1 = getGradF0(X1);
// GradFiX1 = getGradFi(X1);
rPriX1 = rPri(X1);
rCentX1L1t = rCent(fiX1, L1, t);
rDualX1L1V1 = rDual(gradF0X1, L1, V1);
double normRX1L1V1t =
Math.sqrt(ALG.norm2(rPriX1) + ALG.norm2(rCentX1L1t) + ALG.norm2(rDualX1L1V1));
// log.debug("normRX1L1V1t: "+normRX1L1V1t);
if (normRX1L1V1t <= (1 - getAlpha() * s) * normRXLVt) {
break;
}
if (!Double.isNaN(previousNormRX1L1V1t)) {
if (previousNormRX1L1V1t <= normRX1L1V1t) {
log.warn("No progress achieved in backtracking with norm");
break;
}
}
previousNormRX1L1V1t = normRX1L1V1t;
}
s = getBeta() * s;
// log.debug("s: " + s);
}
// update
X = X1;
V = V1;
L = L1;
}
// if(lpRequest.isCheckOptimalDualityConditions()){
// //check duality conditions:
// if(!checkDualityConditions(X, L, V)){
// log.error("duality conditions not satisfied");
// lpResponse.setReturnCode(OptimizationResponse.FAILED);
// throw new Exception("duality conditions not satisfied");
// }
// }
if (lpRequest.isCheckOptimalLagrangianBounds()) {
// check equality constraints Lagrangian bounds
// if(!checkEqConstraintsLagrangianBounds(V)){
// log.error("equality constraints Lagrangian multipliers bounds not satisfied");
// lpResponse.setReturnCode(OptimizationResponse.FAILED);
// throw new Exception("equality constraints Lagrangian multipliers bounds not satisfied");
// }
// check inequality constraints Lagrangian bounds
// if(!checkIneqConstraintsLagrangianBounds(X, L)){
// log.error("inequality constraints Lagrangian multipliers bounds not satisfied");
// lpResponse.setReturnCode(OptimizationResponse.FAILED);
// throw new Exception("inequality constraints Lagrangian multipliers bounds not
// satisfied");
// }
}
long tStop = System.currentTimeMillis();
log.debug("time: " + (tStop - tStart));
log.debug("sol : " + ArrayUtils.toString(X.toArray()));
log.debug("ret code: " + lpResponse.getReturnCode());
// log.debug("XList : " + ArrayUtils.toString(XList));
// for(int s=0; s<SList.size(); s++){
// log.debug("SList : " + ArrayUtils.toString(SList.get(s)));
// }
lpResponse.setSolution(X.toArray());
setLPOptimizationResponse(lpResponse);
return lpResponse.getReturnCode();
}
public String classify(Collection<Rules> cr, DoubleMatrix1D feature) throws GggpExceptionImpl {
boolean evaluation;
Iterator<Rules> itr;
String res = null;
Map<String, Double> aux;
int normalClass;
int injuryClass;
Rules r;
if ((cr != null) && (!cr.isEmpty()) && (feature != null)) {
aux = new ConcurrentHashMap<String, Double>();
/** The features's real values are from 0 to 17. */
aux.put((new String("secDifTorMax").toUpperCase()), feature.get(0));
aux.put((new String("secDifAngTorMax").toUpperCase()), feature.get(1));
aux.put((new String("secDifTorMin").toUpperCase()), feature.get(2));
aux.put((new String("secDifAngTorMin").toUpperCase()), feature.get(3));
aux.put((new String("torMax").toUpperCase()), feature.get(4));
aux.put((new String("angTorMax").toUpperCase()), feature.get(5));
aux.put((new String("timTorMax").toUpperCase()), feature.get(6));
aux.put((new String("torMin").toUpperCase()), feature.get(7));
aux.put((new String("angTorMin").toUpperCase()), feature.get(8));
aux.put((new String("timTorMin").toUpperCase()), feature.get(9));
aux.put((new String("timAvgTorMaxExt").toUpperCase()), feature.get(10));
aux.put((new String("desAvgTimMaxExt").toUpperCase()), feature.get(11));
aux.put((new String("timAvgTorMinFlx").toUpperCase()), feature.get(12));
aux.put((new String("desAvgTimMinFlx").toUpperCase()), feature.get(13));
aux.put((new String("torAvgExt").toUpperCase()), feature.get(14));
aux.put((new String("desAvgTorExt").toUpperCase()), feature.get(15));
aux.put((new String("torAvgFlx").toUpperCase()), feature.get(16));
aux.put((new String("desAvgTorFlx").toUpperCase()), feature.get(17));
itr = cr.iterator();
normalClass = 0;
injuryClass = 0;
while (itr.hasNext()) {
r = itr.next();
evaluation = r.evaluatesRule(aux, r.getAntecedent());
if (evaluation) {
if (((r.getConsecuent()).toUpperCase()).equals("NORMAL")) {
normalClass++;
} else {
injuryClass++;
}
}
}
if (normalClass > injuryClass) {
res = new String("NORMAL");
} else {
if (normalClass < injuryClass) {
res = new String("INJURY");
} else {
res = new String("INDEFINITE");
}
}
}
return res;
}
protected double computeFunctionGradientLL(double lambda[], double grad[]) {
double logli = 0;
try {
for (int f = 0; f < lambda.length; f++) {
grad[f] = -1 * lambda[f] * params.invSigmaSquare;
logli -= ((lambda[f] * lambda[f]) * params.invSigmaSquare) / 2;
}
diter.startScan();
if (featureGenCache != null) featureGenCache.startDataScan();
for (int numRecord = 0; diter.hasNext(); numRecord++) {
DataSequence dataSeq = (DataSequence) diter.next();
if (featureGenCache != null) featureGenCache.nextDataIndex();
if (params.debugLvl > 1) {
Util.printDbg("Read next seq: " + numRecord + " logli " + logli);
}
alpha_Y.assign(0);
for (int f = 0; f < lambda.length; f++) ExpF[f] = RobustMath.LOG0;
if ((beta_Y == null) || (beta_Y.length < dataSeq.length())) {
beta_Y = new DenseDoubleMatrix1D[2 * dataSeq.length()];
for (int i = 0; i < beta_Y.length; i++) beta_Y[i] = new DenseDoubleMatrix1D(numY);
}
// compute beta values in a backward scan.
// also scale beta-values to 1 to avoid numerical problems.
beta_Y[dataSeq.length() - 1].assign(0);
for (int i = dataSeq.length() - 1; i > 0; i--) {
if (params.debugLvl > 2) {
/* Util.printDbg("Features fired");
featureGenerator.startScanFeaturesAt(dataSeq, i);
while (featureGenerator.hasNext()) {
Feature feature = featureGenerator.next();
Util.printDbg(feature.toString());
}
*/
}
// compute the Mi matrix
initMDone =
computeLogMi(
featureGenerator, lambda, dataSeq, i, Mi_YY, Ri_Y, false, reuseM, initMDone);
tmp_Y.assign(beta_Y[i]);
tmp_Y.assign(Ri_Y, sumFunc);
RobustMath.logMult(Mi_YY, tmp_Y, beta_Y[i - 1], 1, 0, false, edgeGen);
}
double thisSeqLogli = 0;
for (int i = 0; i < dataSeq.length(); i++) {
// compute the Mi matrix
initMDone =
computeLogMi(
featureGenerator, lambda, dataSeq, i, Mi_YY, Ri_Y, false, reuseM, initMDone);
// find features that fire at this position..
featureGenerator.startScanFeaturesAt(dataSeq, i);
if (i > 0) {
tmp_Y.assign(alpha_Y);
RobustMath.logMult(Mi_YY, tmp_Y, newAlpha_Y, 1, 0, true, edgeGen);
newAlpha_Y.assign(Ri_Y, sumFunc);
} else {
newAlpha_Y.assign(Ri_Y);
}
while (featureGenerator.hasNext()) {
Feature feature = featureGenerator.next();
int f = feature.index();
int yp = feature.y();
int yprev = feature.yprev();
float val = feature.value();
if ((dataSeq.y(i) == yp)
&& (((i - 1 >= 0) && (yprev == dataSeq.y(i - 1))) || (yprev < 0))) {
grad[f] += val;
thisSeqLogli += val * lambda[f];
if (params.debugLvl > 2) {
System.out.println("Feature fired " + f + " " + feature);
}
}
if (yprev < 0) {
ExpF[f] =
RobustMath.logSumExp(
ExpF[f], newAlpha_Y.get(yp) + RobustMath.log(val) + beta_Y[i].get(yp));
} else {
ExpF[f] =
RobustMath.logSumExp(
ExpF[f],
alpha_Y.get(yprev)
+ Ri_Y.get(yp)
+ Mi_YY.get(yprev, yp)
+ RobustMath.log(val)
+ beta_Y[i].get(yp));
}
}
alpha_Y.assign(newAlpha_Y);
if (params.debugLvl > 2) {
System.out.println("Alpha-i " + alpha_Y.toString());
System.out.println("Ri " + Ri_Y.toString());
System.out.println("Mi " + Mi_YY.toString());
System.out.println("Beta-i " + beta_Y[i].toString());
}
}
double lZx = RobustMath.logSumExp(alpha_Y);
thisSeqLogli -= lZx;
logli += thisSeqLogli;
// update grad.
for (int f = 0; f < grad.length; f++) {
grad[f] -= RobustMath.exp(ExpF[f] - lZx);
}
if (params.debugLvl > 1) {
System.out.println(
"Sequence "
+ thisSeqLogli
+ " logli "
+ logli
+ " log(Zx) "
+ lZx
+ " Zx "
+ Math.exp(lZx));
}
}
if (params.debugLvl > 2) {
for (int f = 0; f < lambda.length; f++) System.out.print(lambda[f] + " ");
System.out.println(" :x");
for (int f = 0; f < lambda.length; f++) System.out.print(grad[f] + " ");
System.out.println(" :g");
}
if (params.debugLvl > 0)
Util.printDbg(
"Iteration "
+ icall
+ " log-likelihood "
+ logli
+ " norm(grad logli) "
+ norm(grad)
+ " norm(x) "
+ norm(lambda));
} catch (Exception e) {
System.out.println("Alpha-i " + alpha_Y.toString());
System.out.println("Ri " + Ri_Y.toString());
System.out.println("Mi " + Mi_YY.toString());
e.printStackTrace();
System.exit(0);
}
return logli;
}
protected double computeFunctionGradient(double lambda[], double grad[]) {
initMDone = false;
if (params.trainerType.equals("ll")) return computeFunctionGradientLL(lambda, grad);
double logli = 0;
try {
for (int f = 0; f < lambda.length; f++) {
grad[f] = -1 * lambda[f] * params.invSigmaSquare;
logli -= ((lambda[f] * lambda[f]) * params.invSigmaSquare) / 2;
}
boolean doScaling = params.doScaling;
diter.startScan();
if (featureGenCache != null) featureGenCache.startDataScan();
int numRecord = 0;
for (numRecord = 0; diter.hasNext(); numRecord++) {
DataSequence dataSeq = (DataSequence) diter.next();
if (featureGenCache != null) featureGenCache.nextDataIndex();
if (params.debugLvl > 1) {
Util.printDbg("Read next seq: " + numRecord + " logli " + logli);
}
alpha_Y.assign(1);
for (int f = 0; f < lambda.length; f++) ExpF[f] = 0;
if ((beta_Y == null) || (beta_Y.length < dataSeq.length())) {
beta_Y = new DenseDoubleMatrix1D[2 * dataSeq.length()];
for (int i = 0; i < beta_Y.length; i++) beta_Y[i] = new DenseDoubleMatrix1D(numY);
scale = new double[2 * dataSeq.length()];
}
// compute beta values in a backward scan.
// also scale beta-values to 1 to avoid numerical problems.
scale[dataSeq.length() - 1] = (doScaling) ? numY : 1;
beta_Y[dataSeq.length() - 1].assign(1.0 / scale[dataSeq.length() - 1]);
for (int i = dataSeq.length() - 1; i > 0; i--) {
if (params.debugLvl > 2) {
Util.printDbg("Features fired");
// featureGenerator.startScanFeaturesAt(dataSeq, i);
// while (featureGenerator.hasNext()) {
// Feature feature = featureGenerator.next();
// Util.printDbg(feature.toString());
// }
}
// compute the Mi matrix
initMDone =
computeLogMi(
featureGenerator, lambda, dataSeq, i, Mi_YY, Ri_Y, true, reuseM, initMDone);
tmp_Y.assign(beta_Y[i]);
tmp_Y.assign(Ri_Y, multFunc);
RobustMath.Mult(Mi_YY, tmp_Y, beta_Y[i - 1], 1, 0, false, edgeGen);
// Mi_YY.zMult(tmp_Y, beta_Y[i-1]);
// need to scale the beta-s to avoid overflow
scale[i - 1] = doScaling ? beta_Y[i - 1].zSum() : 1;
if ((scale[i - 1] < 1) && (scale[i - 1] > -1)) scale[i - 1] = 1;
constMultiplier.multiplicator = 1.0 / scale[i - 1];
beta_Y[i - 1].assign(constMultiplier);
}
double thisSeqLogli = 0;
for (int i = 0; i < dataSeq.length(); i++) {
// compute the Mi matrix
initMDone =
computeLogMi(
featureGenerator, lambda, dataSeq, i, Mi_YY, Ri_Y, true, reuseM, initMDone);
// find features that fire at this position..
featureGenerator.startScanFeaturesAt(dataSeq, i);
if (i > 0) {
tmp_Y.assign(alpha_Y);
RobustMath.Mult(Mi_YY, tmp_Y, newAlpha_Y, 1, 0, true, edgeGen);
// Mi_YY.zMult(tmp_Y, newAlpha_Y,1,0,true);
newAlpha_Y.assign(Ri_Y, multFunc);
} else {
newAlpha_Y.assign(Ri_Y);
}
while (featureGenerator.hasNext()) {
Feature feature = featureGenerator.next();
int f = feature.index();
int yp = feature.y();
int yprev = feature.yprev();
float val = feature.value();
if ((dataSeq.y(i) == yp)
&& (((i - 1 >= 0) && (yprev == dataSeq.y(i - 1))) || (yprev < 0))) {
grad[f] += val;
thisSeqLogli += val * lambda[f];
}
if (yprev < 0) {
ExpF[f] += newAlpha_Y.get(yp) * val * beta_Y[i].get(yp);
} else {
ExpF[f] +=
alpha_Y.get(yprev)
* Ri_Y.get(yp)
* Mi_YY.get(yprev, yp)
* val
* beta_Y[i].get(yp);
}
}
alpha_Y.assign(newAlpha_Y);
// now scale the alpha-s to avoid overflow problems.
constMultiplier.multiplicator = 1.0 / scale[i];
alpha_Y.assign(constMultiplier);
if (params.debugLvl > 2) {
System.out.println("Alpha-i " + alpha_Y.toString());
System.out.println("Ri " + Ri_Y.toString());
System.out.println("Mi " + Mi_YY.toString());
System.out.println("Beta-i " + beta_Y[i].toString());
}
}
double Zx = alpha_Y.zSum();
thisSeqLogli -= log(Zx);
// correct for the fact that alpha-s were scaled.
for (int i = 0; i < dataSeq.length(); i++) {
thisSeqLogli -= log(scale[i]);
}
logli += thisSeqLogli;
// update grad.
for (int f = 0; f < grad.length; f++) grad[f] -= ExpF[f] / Zx;
if (params.debugLvl > 1) {
System.out.println(
"Sequence "
+ thisSeqLogli
+ " logli "
+ logli
+ " log(Zx) "
+ Math.log(Zx)
+ " Zx "
+ Zx);
}
}
if (params.debugLvl > 2) {
for (int f = 0; f < lambda.length; f++) System.out.print(lambda[f] + " ");
System.out.println(" :x");
for (int f = 0; f < lambda.length; f++)
System.out.println(featureGenerator.featureName(f) + " " + grad[f] + " ");
System.out.println(" :g");
}
if (params.debugLvl > 0)
Util.printDbg(
"Iter "
+ icall
+ " log likelihood "
+ logli
+ " norm(grad logli) "
+ norm(grad)
+ " norm(x) "
+ norm(lambda));
if (icall == 0) {
System.out.println("Number of training records" + numRecord);
}
} catch (Exception e) {
System.out.println("Alpha-i " + alpha_Y.toString());
System.out.println("Ri " + Ri_Y.toString());
System.out.println("Mi " + Mi_YY.toString());
e.printStackTrace();
System.exit(0);
}
return logli;
}