/**
* Standard 2D tracking model with the following state equation: {@latex[ D_ x_t = G x_ t-1} + A
* \epsilon_t} Also, when angle != null, a constraint matrix is created for the state covariance,
* with perpendicular variance a0Variance.
*
* @param gVariance
* @param aVariance
* @param a0Variance
* @param angle
*/
public Standard2DTrackingFilter(
double gVariance, double aVariance, double a0Variance, Double angle) {
super(
VectorFactory.getDefault().createVector(4),
createStateCovarianceMatrix(1d, aVariance, a0Variance, angle),
MatrixFactory.getDefault().createIdentity(2, 2).scale(gVariance));
this.aVariance = aVariance;
this.gVariance = gVariance;
this.a0Variance = a0Variance;
this.angle = angle;
final LinearDynamicalSystem model = new LinearDynamicalSystem(0, 4);
final Matrix Gct = createStateTransitionMatrix(currentTimeDiff);
final Matrix G = MatrixFactory.getDefault().createIdentity(4, 4);
G.setSubMatrix(0, 0, Gct);
model.setA(G);
model.setB(MatrixFactory.getDefault().createMatrix(4, 4));
model.setC(O);
this.model = model;
}
private static Matrix createStateTransitionMatrix(double timeDiff) {
final Matrix Gct = MatrixFactory.getDefault().createIdentity(4, 4);
Gct.setElement(0, 1, timeDiff);
Gct.setElement(2, 3, timeDiff);
return Gct;
}
/**
* Evaluates the predictive observation likelihood given the predictive state distribution.
*
* @param obs
* @param belief
* @return
*/
public double predictiveLikelihood(Vector obs, MultivariateGaussian belief) {
Matrix Q = belief.getCovariance();
Q = this.model.getC().times(Q).times(this.model.getC().transpose());
Q.plusEquals(this.measurementCovariance);
final MultivariateGaussian.PDF pdf =
new MultivariateGaussian.PDF(this.model.getC().times(belief.getMean()), Q);
return pdf.evaluate(obs);
}
/**
* Creates a uniform transition-probability Matrix
*
* @param numStates Number of states to create the Matrix for
* @return Uniform probability Matrix.
*/
protected static Matrix createUniformTransitionProbability(int numStates) {
Matrix A = MatrixFactory.getDefault().createMatrix(numStates, numStates);
final double p = 1.0 / numStates;
for (int i = 0; i < numStates; i++) {
for (int j = 0; j < numStates; j++) {
A.setElement(i, j, p);
}
}
return A;
}
public static Matrix plusInplace(Matrix mat, double etat) {
int nrows = mat.getNumRows();
int ncols = mat.getNumColumns();
for (int r = 0; r < nrows; r++) {
for (int c = 0; c < ncols; c++) {
mat.setElement(r, c, mat.getElement(r, c) + etat);
}
}
return mat;
}
public static double colSparcity(Matrix mat) {
double ncols = mat.getNumColumns();
double nsparse = 0;
for (int c = 0; c < ncols; c++) {
if (mat.getColumn(c).sum() == 0) {
nsparse++;
}
}
return nsparse / ncols;
}
public static double rowSparcity(Matrix mat) {
double nrows = mat.getNumRows();
double nsparse = 0;
for (int r = 0; r < nrows; r++) {
if (mat.getRow(r).sum() == 0) {
nsparse++;
}
}
return nsparse / nrows;
}
@Override
public Matrix init(int rows, int cols) {
final Matrix ret = smf.createMatrix(rows, cols);
for (int i = 0; i < rows; i++) {
for (int j = 0; j < cols; j++) {
if (this.random.nextDouble() > sparcity) ret.setElement(i, j, 1d);
}
}
return ret;
}
public static double absSum(Matrix mat) {
double tot = 0;
int nrows = mat.getNumRows();
int ncols = mat.getNumColumns();
for (int r = 0; r < nrows; r++) {
for (int c = 0; c < ncols; c++) {
tot += Math.abs(mat.getElement(r, c));
}
}
return tot;
}
@Override
public Matrix init(int rows, int cols) {
final SparseMatrix rand = (SparseMatrix) smf.createUniformRandom(rows, cols, min, max, random);
final Matrix ret = smf.createMatrix(rows, cols);
for (int i = 0; i < rows; i++) {
if (this.random.nextDouble() > sparcity) {
ret.setRow(i, rand.getRow(i));
}
}
return ret;
}
/**
* Creates a new instance of ContinuousDensityHiddenMarkovModel
*
* @param initialProbability Initial probability Vector over the states. Each entry must be
* nonnegative and the Vector must sum to 1.
* @param transitionProbability Transition probability matrix. The entry (i,j) is the probability
* of transition from state "j" to state "i". As a corollary, all entries in the Matrix must
* be nonnegative and the columns of the Matrix must sum to 1.
*/
public MarkovChain(Vector initialProbability, Matrix transitionProbability) {
if (!transitionProbability.isSquare()) {
throw new IllegalArgumentException("transitionProbability must be square!");
}
final int k = transitionProbability.getNumRows();
initialProbability.assertDimensionalityEquals(k);
this.setTransitionProbability(transitionProbability);
this.setInitialProbability(initialProbability);
}
/**
* Creates a new {@code MatrixBasedTermSimilarityNetwork}.
*
* @param termIndex The index of terms that contains the nodes of the network.
* @param similarities The square matrix of similarities between terms. Must have a number of rows
* and columns equal to the number of terms in the term index.
*/
public MatrixBasedTermSimilarityNetwork(final TermIndex termIndex, final Matrix similarities) {
super();
if (similarities.getNumRows() != termIndex.getTermCount()
|| similarities.getNumColumns() != termIndex.getTermCount()) {
throw new DimensionalityMismatchException(
"the number of terms in the term index must match the "
+ "dimensions of the square similarities matrix");
}
this.setTermIndex(termIndex);
this.setSimilarities(similarities);
}
public static Matrix asMat(MLArray mlArray) {
MLDouble mlArrayDbl = (MLDouble) mlArray;
int rows = mlArray.getM();
int cols = mlArray.getN();
Matrix mat = SparseMatrixFactoryMTJ.INSTANCE.createMatrix(rows, cols);
for (int r = 0; r < rows; r++) {
for (int c = 0; c < cols; c++) {
mat.setElement(r, c, mlArrayDbl.get(r, c));
}
}
return mat;
}
public static Vector diag(Matrix mat) {
Vector ret;
if (mat.getNumColumns() > mat.getNumRows()) {
ret = mat.getRow(0);
} else {
ret = mat.getColumn(0);
}
int rowcol = ret.getDimensionality();
for (int rc = 0; rc < rowcol; rc++) {
ret.setElement(rc, mat.getElement(rc, rc));
}
return ret;
}
/**
* Setter for transitionProbability.
*
* @param transitionProbability Transition probability matrix. The entry (i,j) is the probability
* of transition from state "j" to state "i". As a corollary, all entries in the Matrix must
* be nonnegative and the columns of the Matrix must sum to 1.
*/
public void setTransitionProbability(Matrix transitionProbability) {
if (!transitionProbability.isSquare()) {
throw new IllegalArgumentException("Transition Probability must be square");
}
this.normalizeTransitionMatrix(transitionProbability);
this.transitionProbability = transitionProbability;
}
/**
* Normalizes a column of the transition-probability matrix
*
* @param A Transition probability matrix to normalize, modified by side effect
* @param j Column of the matrix to normalize
*/
protected static void normalizeTransitionMatrix(Matrix A, final int j) {
double sum = 0.0;
final int k = A.getNumRows();
for (int i = 0; i < k; i++) {
final double value = A.getElement(i, j);
if (value < 0.0) {
throw new IllegalArgumentException("Transition Probabilities must be >= 0.0");
}
sum += A.getElement(i, j);
}
if (sum <= 0.0) {
sum = 1.0;
}
if (sum != 1.0) {
for (int i = 0; i < k; i++) {
A.setElement(i, j, A.getElement(i, j) / sum);
}
}
}
@Override
public void measure(MultivariateGaussian belief, Vector observation) {
final Matrix C = this.model.getC();
// Figure out what the model says the observation should be
final Vector xpred = belief.getMean();
final Vector ypred = C.times(xpred);
// Update step... compute the difference between the observation
// and what the model says.
// Then compute the Kalman gain, which essentially indicates
// how much to believe the observation, and how much to believe model
final Vector innovation = observation.minus(ypred);
this.computeMeasurementBelief(belief, innovation, C);
// XXX covariance was set in the previous call
// if (!checkPosDef((DenseMatrix)belief.getCovariance()))
// return;
}
private static Matrix createStateCovarianceMatrix(
double timeDiff, double aVariance, double a0Variance, Double angle) {
final Matrix A_half = MatrixFactory.getDefault().createMatrix(4, 2);
A_half.setElement(0, 0, Math.pow(timeDiff, 2) / 2d);
A_half.setElement(1, 0, timeDiff);
A_half.setElement(2, 1, Math.pow(timeDiff, 2) / 2d);
A_half.setElement(3, 1, timeDiff);
final Matrix Q = getVarianceConstraintMatrix(aVariance, a0Variance, angle);
final Matrix A = A_half.times(Q).times(A_half.transpose());
return A;
}
public double sumLoss(
List<Pair<Matrix>> pairs, Matrix u, Matrix w, Matrix bias, BilinearLearnerParameters params) {
LossFunction loss = params.getTyped(BilinearLearnerParameters.LOSS);
loss = new MatLossFunction(loss);
double total = 0;
int i = 0;
int ntasks = 0;
for (Pair<Matrix> pair : pairs) {
Matrix X = pair.firstObject();
Matrix Y = pair.secondObject();
SparseMatrix Yexp = BilinearSparseOnlineLearner.expandY(Y);
Matrix expectedAll = u.transpose().times(X.transpose()).times(w);
loss.setY(Yexp);
loss.setX(expectedAll);
if (bias != null) loss.setBias(bias);
logger.debug("Testing pair: " + i);
total += loss.eval(null); // Assums an identity w.
i++;
ntasks += Y.getNumColumns();
}
total /= ntasks;
return Math.sqrt(total);
}
public static Matrix abs(Matrix mat) {
Matrix ret = mat.clone();
int nrows = ret.getNumRows();
int ncols = ret.getNumColumns();
for (int r = 0; r < nrows; r++) {
for (int c = 0; c < ncols; c++) {
ret.setElement(r, c, Math.abs(mat.getElement(r, c)));
}
}
return ret;
}
public static Matrix vstack(MatrixFactory<? extends Matrix> matrixFactory, Matrix... matricies) {
int nrows = 0;
int ncols = 0;
for (Matrix matrix : matricies) {
nrows += matrix.getNumRows();
ncols = matrix.getNumColumns();
}
Matrix ret = matrixFactory.createMatrix(nrows, ncols);
int currentRow = 0;
for (Matrix matrix : matricies) {
ret.setSubMatrix(currentRow, 0, matrix);
currentRow += matrix.getNumRows();
}
return ret;
}
@Test
public void testMatlabFile() throws IOException {
MatlabFileDataGenerator gen = new MatlabFileDataGenerator(matfile);
int nusers = -1;
int nwords = -1;
int ntasks = -1;
for (int i = 0; i < gen.size(); i++) {
Pair<Matrix> XY = gen.generate();
Matrix X = XY.firstObject();
Matrix Y = XY.secondObject();
if (nusers == -1) {
nusers = X.getNumColumns();
nwords = X.getNumRows();
ntasks = Y.getNumColumns();
} else {
assertTrue(nusers == X.getNumColumns());
assertTrue(nwords == X.getNumRows());
assertTrue(ntasks == Y.getNumColumns());
}
}
}
private static Matrix getVarianceConstraintMatrix(
double aVariance, double a0Variance, Double angle) {
if (angle == null) return MatrixFactory.getDefault().createIdentity(2, 2).scale(aVariance);
final Matrix rotationMatrix = MatrixFactory.getDefault().createIdentity(2, 2);
rotationMatrix.setElement(0, 0, Math.cos(angle));
rotationMatrix.setElement(0, 1, -Math.sin(angle));
rotationMatrix.setElement(1, 0, Math.sin(angle));
rotationMatrix.setElement(1, 1, Math.cos(angle));
final Matrix temp =
MatrixFactory.getDefault()
.createDiagonal(
VectorFactory.getDefault().copyArray(new double[] {a0Variance, aVariance}));
return rotationMatrix.times(temp).times(rotationMatrix.transpose());
}
/**
* Computes the pair-wise confidence test results
*
* @param data Data from each treatment to consider
* @param pairwiseTest Confidence test used for pair-wise null-hypothesis tests.
*/
protected void computePairwiseTestResults(
final Collection<? extends Collection<? extends Number>> data,
final NullHypothesisEvaluator<Collection<? extends Number>> pairwiseTest) {
ArrayList<? extends Collection<? extends Number>> treatments =
CollectionUtil.asArrayList(data);
final int K = treatments.size();
Matrix Z = MatrixFactory.getDefault().createMatrix(K, K);
Matrix P = MatrixFactory.getDefault().createMatrix(K, K);
ArrayList<ArrayList<ConfidenceStatistic>> stats =
new ArrayList<ArrayList<ConfidenceStatistic>>(K);
for (int i = 0; i < K; i++) {
ArrayList<ConfidenceStatistic> comp = new ArrayList<ConfidenceStatistic>(K);
for (int j = 0; j < K; j++) {
comp.add(null);
}
stats.add(comp);
}
for (int i = 0; i < K; i++) {
// Comparisons to ourselves are perfect
Z.setElement(i, i, 0.0);
P.setElement(i, i, 1.0);
Collection<? extends Number> datai = treatments.get(i);
for (int j = i + 1; j < K; j++) {
Collection<? extends Number> dataj = treatments.get(j);
ConfidenceStatistic comparison = pairwiseTest.evaluateNullHypothesis(datai, dataj);
final double pij = comparison.getNullHypothesisProbability();
final double zij = comparison.getTestStatistic();
Z.setElement(i, j, zij);
Z.setElement(j, i, zij);
P.setElement(i, j, pij);
P.setElement(j, i, pij);
stats.get(i).set(j, comparison);
stats.get(j).set(i, comparison);
}
}
this.testStatistics = Z;
this.nullHypothesisProbabilities = P;
this.pairwiseTestStatistics = stats;
}
static {
O = MatrixFactory.getDefault().createMatrix(2, 4);
O.setElement(0, 0, 1);
O.setElement(1, 2, 1);
}
// Note this method isn't static, because we want to override to make
// it parallelized --krdixon
protected void normalizeTransitionMatrix(Matrix A) {
final int k = A.getNumColumns();
for (int j = 0; j < k; j++) {
normalizeTransitionMatrix(A, j);
}
}