Package org.apache.mahout.math.function

Examples of org.apache.mahout.math.function.PlusMult

556
557
558
559
560
561
562
563
564
565
566
    Vector w = new DenseVector(columns);
    for (int i = 0; i < rows; i++) {
      Vector xi = viewRow(i);
      double d = xi.dot(v);
      if (d != 0.0) {
        w.assign(xi, new PlusMult(d));
      }

    }
    return w;
  }
View Full Code Here
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
    if (numPreviousEigens > 0 && currentState.isFirstPass()) {
      updateTrainingProjectionsVector(currentState, trainingVector, numPreviousEigens - 1);
    }
    if (currentState.getActivationDenominatorSquared() == 0 || trainingVectorNorm == 0) {
      if (currentState.getActivationDenominatorSquared() == 0) {
        pseudoEigen.assign(trainingVector, new PlusMult(1));
        currentState.setHelperVector(currentState.currentTrainingProjection().clone());
        double helperNorm = currentState.getHelperVector().norm(2);
        currentState.setActivationDenominatorSquared(trainingVectorNorm * trainingVectorNorm - helperNorm * helperNorm);
      }
      return;
    }
    currentState.setActivationNumerator(pseudoEigen.dot(trainingVector));
    currentState.setActivationNumerator(
        currentState.getActivationNumerator()
            - currentState.getHelperVector().dot(currentState.currentTrainingProjection()));

    double activation = currentState.getActivationNumerator()
        / Math.sqrt(currentState.getActivationDenominatorSquared());
    currentState.setActivationDenominatorSquared(
        currentState.getActivationDenominatorSquared()
            + 2 * activation * currentState.getActivationNumerator()
            + activation * activation
                * (trainingVector.getLengthSquared() - currentState.currentTrainingProjection().getLengthSquared()));
    if (numPreviousEigens > 0) {
      currentState.getHelperVector().assign(currentState.currentTrainingProjection(), new PlusMult(activation));
    }
    pseudoEigen.assign(trainingVector, new PlusMult(activation));
  }
View Full Code Here
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
      if (state.getScaleFactor() <= 0) {
        state.setScaleFactor(calculateScaleFactor(nextVector));
      }
      nextVector.assign(new Scale(1.0 / state.getScaleFactor()));
      if (previousVector != null) {
        nextVector.assign(previousVector, new PlusMult(-beta));
      }
      // now orthogonalize
      double alpha = currentVector.dot(nextVector);
      nextVector.assign(currentVector, new PlusMult(-alpha));
      endTime(TimingSection.ITERATE);
      startTime(TimingSection.ORTHOGANLIZE);
      orthoganalizeAgainstAllButLast(nextVector, state);
      endTime(TimingSection.ORTHOGANLIZE);
      // and normalize
      beta = nextVector.norm(2);
      if (outOfRange(beta) || outOfRange(alpha)) {
        log.warn("Lanczos parameters out of range: alpha = {}, beta = {}.  Bailing out early!",
            alpha, beta);
        break;
      }
      nextVector.assign(new Scale(1 / beta));
      state.setBasisVector(i, nextVector);
      previousVector = currentVector;
      currentVector = nextVector;
      // save the projections and norms!
      triDiag.set(i - 1, i - 1, alpha);
      if (i < desiredRank - 1) {
        triDiag.set(i - 1, i, beta);
        triDiag.set(i, i - 1, beta);
      }
      state.setIterationNumber(++i);
    }
    startTime(TimingSection.TRIDIAG_DECOMP);

    log.info("Lanczos iteration complete - now to diagonalize the tri-diagonal auxiliary matrix.");
    // at this point, have tridiag all filled out, and basis is all filled out, and orthonormalized
    EigenDecomposition decomp = new EigenDecomposition(triDiag);

    Matrix eigenVects = decomp.getV();
    Vector eigenVals = decomp.getRealEigenvalues();
    endTime(TimingSection.TRIDIAG_DECOMP);
    startTime(TimingSection.FINAL_EIGEN_CREATE);
    for (int row = 0; row < i; row++) {
      Vector realEigen = null;
      // the eigenvectors live as columns of V, in reverse order.  Weird but true.
      Vector ejCol = eigenVects.viewColumn(i - row - 1);
      int size = Math.min(ejCol.size(), state.getBasisSize());
      for (int j = 0; j < size; j++) {
        double d = ejCol.get(j);
        Vector rowJ = state.getBasisVector(j);
        if (realEigen == null) {
          realEigen = rowJ.like();
        }
        realEigen.assign(rowJ, new PlusMult(d));
      }

      Preconditions.checkState(realEigen != null);
      assert realEigen != null;
View Full Code Here
195
196
197
198
199
200
201
202
      Vector basisVector = state.getBasisVector(i);
      double alpha;
      if (basisVector == null || (alpha = nextVector.dot(basisVector)) == 0.0) {
        continue;
      }
      nextVector.assign(basisVector, new PlusMult(-alpha));
    }
  }
View Full Code Here
273
274
275
276
277
278
279
280
281
282
283
     * Step 1: orthogonalize currentPseudoEigen by subtracting off eigen(i) * helper.get(i)
     * Step 2: zero-out the helper vector because it has already helped.
     */
    for (int i = 0; i < state.getNumEigensProcessed(); i++) {
      Vector previousEigen = previousEigens.viewRow(i);
      currentPseudoEigen.assign(previousEigen, new PlusMult(-state.getHelperVector().get(i)));
      state.getHelperVector().set(i, 0);
    }
    if (DEBUG && currentPseudoEigen.norm(2) > 0) {
      for (int i = 0; i < state.getNumEigensProcessed(); i++) {
        Vector previousEigen = previousEigens.viewRow(i);
View Full Code Here
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
      if(state.getScaleFactor() <= 0) {
        state.setScaleFactor(calculateScaleFactor(nextVector));
      }
      nextVector.assign(new Scale(1.0 / state.getScaleFactor()));
      if(previousVector != null) {
        nextVector.assign(previousVector, new PlusMult(-beta));
      }
      // now orthogonalize
      double alpha = currentVector.dot(nextVector);
      nextVector.assign(currentVector, new PlusMult(-alpha));
      endTime(TimingSection.ITERATE);
      startTime(TimingSection.ORTHOGANLIZE);
      orthoganalizeAgainstAllButLast(nextVector, state);
      endTime(TimingSection.ORTHOGANLIZE);
      // and normalize
      beta = nextVector.norm(2);
      if (outOfRange(beta) || outOfRange(alpha)) {
        log.warn("Lanczos parameters out of range: alpha = {}, beta = {}.  Bailing out early!",
            alpha, beta);
        break;
      }
      nextVector.assign(new Scale(1 / beta));
      state.setBasisVector(i, nextVector);
      previousVector = currentVector;
      currentVector = nextVector;
      // save the projections and norms!
      triDiag.set(i - 1, i - 1, alpha);
      if (i < desiredRank - 1) {
        triDiag.set(i - 1, i, beta);
        triDiag.set(i, i - 1, beta);
      }
      state.setIterationNumber(++i);
    }
    startTime(TimingSection.TRIDIAG_DECOMP);

    log.info("Lanczos iteration complete - now to diagonalize the tri-diagonal auxiliary matrix.");
    // at this point, have tridiag all filled out, and basis is all filled out, and orthonormalized
    EigenvalueDecomposition decomp = new EigenvalueDecomposition(triDiag);

    DoubleMatrix2D eigenVects = decomp.getV();
    DoubleMatrix1D eigenVals = decomp.getRealEigenvalues();
    endTime(TimingSection.TRIDIAG_DECOMP);
    startTime(TimingSection.FINAL_EIGEN_CREATE);
    for (int row = 0; row < i; row++) {
      Vector realEigen = null;
      // the eigenvectors live as columns of V, in reverse order.  Weird but true.
      DoubleMatrix1D ejCol = eigenVects.viewColumn(i - row - 1);
      int size = ejCol.size();
      for (int j = 0; j < size; j++) {
        double d = ejCol.get(j);
        Vector rowJ = state.getBasisVector(j);
        if(realEigen == null) {
          realEigen = rowJ.like();
        }
        realEigen.assign(rowJ, new PlusMult(d));
      }
      realEigen = realEigen.normalize();
      state.setRightSingularVector(row, realEigen);
      double e = eigenVals.get(row) * state.getScaleFactor();
      if(!isSymmetric) {
View Full Code Here
185
186
187
188
189
190
191
192
      Vector basisVector = state.getBasisVector(i);
      double alpha;
      if (basisVector == null || (alpha = nextVector.dot(basisVector)) == 0.0) {
        continue;
      }
      nextVector.assign(basisVector, new PlusMult(-alpha));
    }
  }
View Full Code Here
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
    final DoubleMatrix1D[] Crows = new DoubleMatrix1D[m];
    for (int i = m; --i >= 0;) {
      Crows[i] = C.viewRow(i);
    }

    final PlusMult fun = PlusMult.plusMult(0);

    this.elements.forEachPair(
        new IntDoubleProcedure() {
          @Override
          public boolean apply(int key, double value) {
            int i = key / columns;
            int j = key % columns;
            fun.setMultiplicator(value * alpha);
            if (transposeA) {
              Crows[j].assign(Brows[i], fun);
            } else {
              Crows[i].assign(Brows[j], fun);
            }
View Full Code Here
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
            numPreviousEigens - 1);
      }
    }
    if (currentState.getActivationDenominatorSquared() == 0 || trainingVectorNorm == 0) {
      if (currentState.getActivationDenominatorSquared() == 0) {
        pseudoEigen.assign(trainingVector, new PlusMult(1));
        currentState.setHelperVector(currentState.currentTrainingProjection().clone());
        double helperNorm = currentState.getHelperVector().norm(2);
        currentState.setActivationDenominatorSquared(trainingVectorNorm * trainingVectorNorm - helperNorm * helperNorm);
      }
      return;
    }
    currentState.setActivationNumerator(pseudoEigen.dot(trainingVector));
    currentState.setActivationNumerator(
        currentState.getActivationNumerator()
            - currentState.getHelperVector().dot(currentState.currentTrainingProjection()));

    double activation = currentState.getActivationNumerator() / Math.sqrt(currentState.getActivationDenominatorSquared());
    currentState.setActivationDenominatorSquared(
        currentState.getActivationDenominatorSquared()
            + 2 * activation * currentState.getActivationNumerator()
            + activation * activation * (trainingVector.getLengthSquared() - currentState.currentTrainingProjection().getLengthSquared()));
    if (numPreviousEigens > 0) {
      currentState.getHelperVector().assign(currentState.currentTrainingProjection(), new PlusMult(activation));
    }
    pseudoEigen.assign(trainingVector, new PlusMult(activation));
  }
View Full Code Here
275
276
277
278
279
280
281
282
283
284
285
     * Step 1: orthogonalize currentPseudoEigen by subtracting off eigen(i) * helper.get(i)
     * Step 2: zero-out the helper vector because it has already helped.
     */
    for (int i = 0; i < state.getNumEigensProcessed(); i++) {
      Vector previousEigen = previousEigens.getRow(i);
      currentPseudoEigen.assign(previousEigen, new PlusMult(-state.getHelperVector().get(i)));
      state.getHelperVector().set(i, 0);
    }
    if (DEBUG && currentPseudoEigen.norm(2) > 0) {
      for (int i = 0; i < state.getNumEigensProcessed(); i++) {
        Vector previousEigen = previousEigens.getRow(i);
View Full Code Here
TOP

Related Classes of org.apache.mahout.math.function.PlusMult

  • org.apache.mahout.math.AbstractMatrix
  • org.apache.mahout.math.decomposer.hebbian.HebbianSolver
  • org.apache.mahout.math.decomposer.hebbian.HebbianUpdater
  • org.apache.mahout.math.decomposer.lanczos.LanczosSolver
  • org.apache.mahout.math.matrix.impl.RCDoubleMatrix2D
  • org.apache.mahout.math.matrix.impl.SparseDoubleMatrix2D
  • org.apache.mahout.math.matrix.impl.TridiagonalDoubleMatrix2D
  • org.apache.mahout.math.matrix.linalg.LUDecompositionQuick
  • org.apache.mahout.math.matrix.linalg.SeqBlas
    • Copyright © 2018 www.massapicom. All rights reserved.
    All source code are property of their respective owners. Java is a trademark of Sun Microsystems, Inc and owned by ORACLE Inc. Contact coftware#gmail.com.