// Makes and scales the matrices V, D, and VT (to avoid ugly decimals)
private void makeVDVT(EigenDecomp ed) {
V = ed.getV();
D = ed.getD();
VT = ed.getVT();
double ref = 0;
for (int i = 0; i < V.getRowDimension(); i++) {
ref = 0;
for (int j = 0; j < V.getColumnDimension(); j++) {
if (V.getEntry(j, i) != 0 && ref == 0) {
ref = V.getEntry(j, i);
}
if (ref != 0) {
V.setEntry(j, i, V.getEntry(j, i) / Math.abs(ref));
}
}
}
for (int i = 0; i < VT.getRowDimension(); i++) {
ref = 0;
for (int j = 0; j < VT.getColumnDimension(); j++) {
if (VT.getEntry(j, i) != 0 && ref == 0) {
ref = VT.getEntry(j, i);
}
if (ref != 0) {
VT.setEntry(j, i, VT.getEntry(j, i) / Math.abs(ref));
}
}
}
}
/**
* Computes the Kendall's Tau rank correlation matrix for the columns of the input matrix.
*
* @param matrix matrix with columns representing variables to correlate
* @return correlation matrix
*/
public RealMatrix computeCorrelationMatrix(final RealMatrix matrix) {
int nVars = matrix.getColumnDimension();
RealMatrix outMatrix = new BlockRealMatrix(nVars, nVars);
for (int i = 0; i < nVars; i++) {
for (int j = 0; j < i; j++) {
double corr = correlation(matrix.getColumn(i), matrix.getColumn(j));
outMatrix.setEntry(i, j, corr);
outMatrix.setEntry(j, i, corr);
}
outMatrix.setEntry(i, i, 1d);
}
return outMatrix;
}
/**
* Compute a covariance matrix from a matrix whose columns represent covariates.
*
* @param matrix input matrix (must have at least one column and two rows)
* @param biasCorrected determines whether or not covariance estimates are bias-corrected
* @return covariance matrix
* @throws MathIllegalArgumentException if the matrix does not contain sufficient data
*/
protected RealMatrix computeCovarianceMatrix(RealMatrix matrix, boolean biasCorrected)
throws MathIllegalArgumentException {
int dimension = matrix.getColumnDimension();
Variance variance = new Variance(biasCorrected);
RealMatrix outMatrix = new BlockRealMatrix(dimension, dimension);
for (int i = 0; i < dimension; i++) {
for (int j = 0; j < i; j++) {
double cov = covariance(matrix.getColumn(i), matrix.getColumn(j), biasCorrected);
outMatrix.setEntry(i, j, cov);
outMatrix.setEntry(j, i, cov);
}
outMatrix.setEntry(i, i, variance.evaluate(matrix.getColumn(i)));
}
return outMatrix;
}
/**
* Derives a correlation matrix from a covariance matrix.
*
* <p>Uses the formula <br>
* <code>r(X,Y) = cov(X,Y)/s(X)s(Y)</code> where <code>r(·,·)</code> is the
* correlation coefficient and <code>s(·)</code> means standard deviation.
*
* @param covarianceMatrix the covariance matrix
* @return correlation matrix
*/
public RealMatrix covarianceToCorrelation(RealMatrix covarianceMatrix) {
int nVars = covarianceMatrix.getColumnDimension();
RealMatrix outMatrix = new BlockRealMatrix(nVars, nVars);
for (int i = 0; i < nVars; i++) {
double sigma = FastMath.sqrt(covarianceMatrix.getEntry(i, i));
outMatrix.setEntry(i, i, 1d);
for (int j = 0; j < i; j++) {
double entry =
covarianceMatrix.getEntry(i, j)
/ (sigma * FastMath.sqrt(covarianceMatrix.getEntry(j, j)));
outMatrix.setEntry(i, j, entry);
outMatrix.setEntry(j, i, entry);
}
}
return outMatrix;
}
/**
* Build a rating matrix from the rating data. Each user's ratings are first normalized by
* subtracting a baseline score (usually a mean).
*
* @param userMapping The index mapping of user IDs to column numbers.
* @param itemMapping The index mapping of item IDs to row numbers.
* @return A matrix storing the <i>normalized</i> user ratings.
*/
private RealMatrix createRatingMatrix(IdIndexMapping userMapping, IdIndexMapping itemMapping) {
final int nusers = userMapping.size();
final int nitems = itemMapping.size();
// Create a matrix with users on rows and items on columns
logger.info("creating {} by {} rating matrix", nusers, nitems);
RealMatrix matrix = MatrixUtils.createRealMatrix(nusers, nitems);
// populate it with data
Cursor<UserHistory<Event>> users = userEventDAO.streamEventsByUser();
try {
for (UserHistory<Event> user : users) {
// Get the row number for this user
int u = userMapping.getIndex(user.getUserId());
MutableSparseVector ratings = Ratings.userRatingVector(user.filter(Rating.class));
MutableSparseVector baselines = MutableSparseVector.create(ratings.keySet());
baselineScorer.score(user.getUserId(), baselines);
// TODO Populate this user's row with their ratings, minus the baseline scores
for (VectorEntry entry : ratings.fast(State.SET)) {
long itemid = entry.getKey();
int i = itemMapping.getIndex(itemid);
double rating = entry.getValue();
double baseline = baselines.get(itemid);
matrix.setEntry(u, i, rating - baseline);
}
}
} finally {
users.close();
}
return matrix;
}
/**
* Calculates {@code P(D_n < d)} using method described in [1] and doubles (see above).
*
* @param d statistic
* @return the two-sided probability of {@code P(D_n < d)}
* @throws MathArithmeticException if algorithm fails to convert {@code h} to a {@link
* org.apache.commons.math3.fraction.BigFraction} in expressing {@code d} as {@code (k - h) /
* m} for integer {@code k, m} and {@code 0 <= h < 1}.
*/
private double roundedK(double d) throws MathArithmeticException {
final int k = (int) FastMath.ceil(n * d);
final FieldMatrix<BigFraction> HBigFraction = this.createH(d);
final int m = HBigFraction.getRowDimension();
/*
* Here the rounding part comes into play: use
* RealMatrix instead of FieldMatrix<BigFraction>
*/
final RealMatrix H = new Array2DRowRealMatrix(m, m);
for (int i = 0; i < m; ++i) {
for (int j = 0; j < m; ++j) {
H.setEntry(i, j, HBigFraction.getEntry(i, j).doubleValue());
}
}
final RealMatrix Hpower = H.power(n);
double pFrac = Hpower.getEntry(k - 1, k - 1);
for (int i = 1; i <= n; ++i) {
pFrac *= (double) i / (double) n;
}
return pFrac;
}
public static RealMatrix getRealMatrixFromJamaMatrix(Matrix m) {
final int rDim = m.getRowDimension();
final int cDim = m.getColumnDimension();
RealMatrix rm = new Array2DRowRealMatrix(rDim, cDim);
for (int i = 0; i < rDim; i++) {
for (int j = 0; j < cDim; j++) {
rm.setEntry(i, j, m.get(i, j));
}
}
return rm;
}
@Override
public void train(List<Instance> instances) {
// ------------------------ initialize rows and columns ---------------------
int rows = instances.size();
int columns = 0;
// get max columns
for (Instance i : instances) {
int localColumns = Collections.max(i.getFeatureVector().getFeatureMap().keySet());
if (localColumns > columns) columns = localColumns;
}
// ------------------------ initialize alpha vector -----------------------
alpha = new ArrayRealVector(rows, 0);
// ------------------------ initialize base X and Y for use --------------------------
double[][] X = new double[rows][columns];
double[] Y = new double[rows];
for (int i = 0; i < rows; i++) {
Y[i] = ((ClassificationLabel) instances.get(i).getLabel()).getLabelValue();
for (int j = 0; j < columns; j++) {
X[i][j] = instances.get(i).getFeatureVector().get(j + 1);
}
}
// ---------------------- gram matrix -------------------
matrixX = new Array2DRowRealMatrix(X);
RealMatrix gram = new Array2DRowRealMatrix(rows, rows);
for (int i = 0; i < rows; i++) {
for (int j = 0; j < rows; j++) {
gram.setEntry(i, j, kernelFunction(matrixX.getRowVector(i), matrixX.getRowVector(j)));
}
}
// ---------------------- gradient ascent --------------------------
Sigmoid g = new Sigmoid(); // helper function
System.out.println("Training start...");
System.out.println(
"Learning rate: " + _learning_rate + " Training times: " + _training_iterations);
for (int idx = 0; idx < _training_iterations; idx++) {
System.out.println("Training iteration: " + (idx + 1));
for (int k = 0; k < rows; k++) {
double gradient_ascent = 0.0;
RealVector alpha_gram = gram.operate(alpha);
for (int i = 0; i < rows; i++) {
double lambda = alpha_gram.getEntry(i);
double kernel = gram.getEntry(i, k);
gradient_ascent =
gradient_ascent
+ Y[i] * g.value(-lambda) * kernel
+ (1 - Y[i]) * g.value(lambda) * (-kernel);
}
alpha.setEntry(k, alpha.getEntry(k) + _learning_rate * gradient_ascent);
}
}
System.out.println("Training done!");
}
private RealMatrix normalizeData(RealMatrix matrix, UserProfileEigenModel model) {
RealMatrix normalizedData =
new Array2DRowRealMatrix(matrix.getRowDimension(), matrix.getColumnDimension());
if (LOG.isDebugEnabled()) LOG.debug("model statistics size: " + model.statistics().length);
for (int i = 0; i < matrix.getRowDimension(); i++) {
for (int j = 0; j < matrix.getColumnDimension(); j++) {
double value =
(matrix.getEntry(i, j) - model.statistics()[j].getMean())
/ model.statistics()[j].getStddev();
normalizedData.setEntry(i, j, value);
}
}
return normalizedData;
}
private RealMatrix removeZeroColumns(RealMatrix base, List<Integer> zeroColumns) {
int adjustedDim = base.getRowDimension() - zeroColumns.size();
if (adjustedDim == 0) return base;
RealMatrix adjusted = new Array2DRowRealMatrix(adjustedDim, adjustedDim);
int i = 0, j = 0;
for (int basei = 0; basei < base.getRowDimension(); basei++) {
if (zeroColumns.contains(basei)) continue;
for (int basej = 0; basej < base.getColumnDimension(); basej++) {
if (zeroColumns.contains(basej)) continue;
adjusted.setEntry(i, j++, base.getEntry(basei, basej));
}
i++;
j = 0;
}
return adjusted;
}
@Override
public RealMatrix computeIOCorrelationMatrix(SampleIterator it) {
it.reset();
Sample sample = it.next();
int inputDim = sample.getEncodedInput().getDimension();
int outputDim = sample.getEncodedOutput().getDimension();
RealMatrix M = new Array2DRowRealMatrix(inputDim, outputDim);
for (int i = 0; i < inputDim; i++) {
for (int j = 0; j < outputDim; j++) {
it.reset();
M.setEntry(i, j, correlationRatio(it, i, j));
}
}
return M;
}
public double valueAt(double[] param) {
double[] sdInv = new double[nVariables];
for (int i = 0; i < nVariables; i++) {
R.setEntry(i, i, 1.0 - param[i]);
sdInv[i] = 1.0 / Sinv.getEntry(i, i);
}
DiagonalMatrix diagSdInv = new DiagonalMatrix(sdInv);
EigenDecomposition eigen = new EigenDecomposition(R);
RealMatrix eigenVectors = eigen.getV().getSubMatrix(0, nVariables - 1, 0, nFactors - 1);
double[] ev = new double[nFactors];
for (int i = 0; i < nFactors; i++) {
ev[i] = Math.sqrt(eigen.getRealEigenvalue(i));
}
DiagonalMatrix evMatrix =
new DiagonalMatrix(
ev); // USE Apache version of Diagonal matrix when upgrade to version 3.2
RealMatrix LAMBDA = eigenVectors.multiply(evMatrix);
RealMatrix SIGMA = (LAMBDA.multiply(LAMBDA.transpose()));
double value = 0.0;
RealMatrix DIF = R.subtract(SIGMA);
for (int i = 0; i < DIF.getRowDimension(); i++) {
for (int j = 0; j < DIF.getColumnDimension(); j++) {
value = DIF.getEntry(i, j);
DIF.setEntry(i, j, Math.pow(value, 2));
}
}
RealMatrix RESID = diagSdInv.multiply(DIF).multiply(diagSdInv);
double sum = 0.0;
for (int i = 0; i < RESID.getRowDimension(); i++) {
for (int j = 0; j < RESID.getColumnDimension(); j++) {
sum += RESID.getEntry(i, j);
}
}
return sum;
}
@Test
public void testSubtractMatrixNorm() {
RealMatrix m, mp;
m = MatrixUtils.createRealMatrix(3, 5);
mp = MatrixUtils.createRealMatrix(3, 5);
m.setEntry(1, 2, 0.1613351);
mp.setEntry(1, 2, 0.486433333333);
Assert.assertEquals(0.0, Graphulo.nmfError_Client(m, mp), 0.000001);
m.setEntry(2, 2, 0.76);
mp.setEntry(2, 2, 0.25);
Assert.assertEquals(0.0, Graphulo.nmfError_Client(m, mp), 0.000001);
m.setEntry(2, 1, 0.4);
mp.setEntry(2, 0, 0.9);
Assert.assertEquals(2.0 / 3.0, Graphulo.nmfError_Client(m, mp), 0.000001);
}
@Override
public List<MLCallbackResult> detect(
final String user,
final String algorithm,
UserActivityAggModel userActivity,
UserProfileEigenModel aModel) {
RealMatrix inputData = userActivity.matrix();
LOG.warn(
"EigenBasedAnomalyDetection predictAnomaly called with dimension: "
+ inputData.getRowDimension()
+ "x"
+ inputData.getColumnDimension());
if (aModel == null) {
LOG.warn(
"nothing to do as the input model does not have required values, returning from evaluating this algorithm..");
return null;
}
List<MLCallbackResult> mlCallbackResults = new ArrayList<MLCallbackResult>();
RealMatrix normalizedMat = normalizeData(inputData, aModel);
UserCommandStatistics[] listStats = aModel.statistics();
int colWithHighVariant = 0;
for (int j = 0; j < normalizedMat.getColumnDimension(); j++) {
if (listStats[j].isLowVariant() == false) {
colWithHighVariant++;
}
}
final Map<String, String> context =
new HashMap<String, String>() {
{
put(UserProfileConstants.USER_TAG, user);
put(UserProfileConstants.ALGORITHM_TAG, algorithm);
}
};
Map<Integer, String> lineNoWithVariantBasedAnomalyDetection = new HashMap<Integer, String>();
for (int i = 0; i < normalizedMat.getRowDimension(); i++) {
MLCallbackResult aResult = new MLCallbackResult();
aResult.setContext(context);
for (int j = 0; j < normalizedMat.getColumnDimension(); j++) {
// LOG.info("mean for j=" + j + " is:" + listStats[j].getMean());
// LOG.info("stddev for j=" + j + " is:" + listStats[j].getStddev());
if (listStats[j].isLowVariant() == true) {
// LOG.info(listOfCmds[j] + " is low variant");
if (normalizedMat.getEntry(i, j) > listStats[j].getMean()) {
lineNoWithVariantBasedAnomalyDetection.put(i, "lowVariantAnomaly");
aResult.setAnomaly(true);
aResult.setTimestamp(userActivity.timestamp());
aResult.setFeature(listStats[j].getCommandName());
aResult.setAlgorithm(UserProfileConstants.EIGEN_DECOMPOSITION_ALGORITHM);
List<String> datapoints = new ArrayList<String>();
double[] rowVals = inputData.getRow(i);
for (double rowVal : rowVals) datapoints.add(rowVal + "");
aResult.setDatapoints(datapoints);
aResult.setId(user);
mlCallbackResults.add(aResult);
} else {
aResult.setAnomaly(false);
aResult.setTimestamp(userActivity.timestamp());
mlCallbackResults.add(aResult);
}
}
}
// return results;
}
// LOG.info("results size here: " + results.length);
// LOG.info("col with high variant: " + colWithHighVariant);
RealMatrix finalMatWithoutLowVariantFeatures =
new Array2DRowRealMatrix(normalizedMat.getRowDimension(), colWithHighVariant);
// LOG.info("size of final test data: " + finalMatWithoutLowVariantFeatures.getRowDimension()
// +"x"+ finalMatWithoutLowVariantFeatures.getColumnDimension());
int finalMatrixRow = 0;
int finalMatrixCol = 0;
for (int i = 0; i < normalizedMat.getRowDimension(); i++) {
for (int j = 0; j < normalizedMat.getColumnDimension(); j++) {
if (listStats[j].isLowVariant() == false) {
finalMatWithoutLowVariantFeatures.setEntry(
finalMatrixRow, finalMatrixCol, normalizedMat.getEntry(i, j));
finalMatrixCol++;
}
}
finalMatrixCol = 0;
finalMatrixRow++;
}
RealVector[] pcs = aModel.principalComponents();
// LOG.info("pc size: " + pcs.getRowDimension() +"x" + pcs.getColumnDimension());
RealMatrix finalInputMatTranspose = finalMatWithoutLowVariantFeatures.transpose();
for (int i = 0; i < finalMatWithoutLowVariantFeatures.getRowDimension(); i++) {
if (lineNoWithVariantBasedAnomalyDetection.get(i) == null) {
MLCallbackResult result = new MLCallbackResult();
result.setContext(context);
for (int sz = 0; sz < pcs.length; sz++) {
double[] pc1 = pcs[sz].toArray();
RealMatrix pc1Mat = new Array2DRowRealMatrix(pc1);
RealMatrix transposePC1Mat = pc1Mat.transpose();
RealMatrix testData =
pc1Mat.multiply(transposePC1Mat).multiply(finalInputMatTranspose.getColumnMatrix(i));
// LOG.info("testData size: " + testData.getRowDimension() + "x" +
// testData.getColumnDimension());
RealMatrix testDataTranspose = testData.transpose();
// LOG.info("testData transpose size: " + testDataTranspose.getRowDimension() + "x" +
// testDataTranspose.getColumnDimension());
RealVector iRowVector = testDataTranspose.getRowVector(0);
// RealVector pc1Vector = transposePC1Mat.getRowVector(sz);
RealVector pc1Vector = transposePC1Mat.getRowVector(0);
double distanceiRowAndPC1 = iRowVector.getDistance(pc1Vector);
// LOG.info("distance from pc sz: " + sz + " " + distanceiRowAndPC1 + " " +
// model.getMaxL2Norm().getEntry(sz));
// LOG.info("model.getMaxL2Norm().getEntry(sz):" + model.getMaxL2Norm().getEntry(sz));
if (distanceiRowAndPC1 > aModel.maximumL2Norm().getEntry(sz)) {
// LOG.info("distance from pc sz: " + sz + " " + distanceiRowAndPC1 + " " +
// model.getMaxL2Norm().getEntry(sz));
result.setAnomaly(true);
result.setFeature(aModel.statistics()[sz].getCommandName());
result.setTimestamp(System.currentTimeMillis());
result.setAlgorithm(UserProfileConstants.EIGEN_DECOMPOSITION_ALGORITHM);
List<String> datapoints = new ArrayList<String>();
double[] rowVals = inputData.getRow(i);
for (double rowVal : rowVals) datapoints.add(rowVal + "");
result.setDatapoints(datapoints);
result.setId(user);
}
}
mlCallbackResults.add(result);
}
}
return mlCallbackResults;
}
