Package de.jungblut.math.DoubleVector

Examples of de.jungblut.math.DoubleVector.DoubleVectorElement

94
95
96
97
98
99
100
101
102
103
104
105
      SparseDoubleVector tmp = new SparseDoubleVector(
          features.getDimension() + 1);
      tmp.set(0, 1d);
      Iterator<DoubleVectorElement> iterateNonZero = features.iterateNonZero();
      while (iterateNonZero.hasNext()) {
        DoubleVectorElement next = iterateNonZero.next();
        tmp.set(next.getIndex() + 1, next.getValue());
      }
      features = tmp;
    } else {
      features = new DenseDoubleVector(1d, features.toArray());
    }
View Full Code Here
389
390
391
392
393
394
395
396
397
398
399
400
401
      com.google.common.hash.HashFunction hashFunction) {
    DoubleVector dense = new DenseDoubleVector(n);
    Iterator<DoubleVectorElement> iterateNonZero = inputFeature
        .iterateNonZero();
    while (iterateNonZero.hasNext()) {
      DoubleVectorElement next = iterateNonZero.next();
      // get the hash as int
      int hash = hashFunction.hashInt(next.getIndex()).asInt();
      // abs it, as we don't want to have negative index access
      int bucket = Math.abs(hash) % n;
      // subtract 1 in case of negative values, else increment.
      // this replaces the second hash function proposed by Weinberger et al.
      dense.set(bucket, dense.get(bucket) + (hash < 0 ? -1d : 1d));
View Full Code Here
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
        DoubleVector activationDifference = activations.subtract(outcome);
        // update theta by a smarter sparsity algorithm
        Iterator<DoubleVectorElement> featureIterator = feature
            .iterateNonZero();
        while (featureIterator.hasNext()) {
          DoubleVectorElement next = featureIterator.next();
          DoubleVector rowVector = theta.getRowVector(next.getIndex());
          double l2 = rowVector.pow(2d).sum();
          Iterator<DoubleVectorElement> diffIterator = activationDifference
              .iterateNonZero();
          while (diffIterator.hasNext()) {
            DoubleVectorElement diffElement = diffIterator.next();
            double val = rowVector.get(diffElement.getIndex());
            if (val != 0) {
              val = val - diffElement.getValue() * alpha;
              // apply the decay
              if (lambda != 0d) {
                val -= ((lambda * val) + (1d - lambda) * l2);
              }
              if (Math.abs(val) < nearZeroLimit) {
                val = 0;
              }
              rowVector.set(diffElement.getIndex(), val);
            }
          }
        }
        if (verbose && localItems % reportInterval == 0) {
          System.out.format(" Item %d | AVG Loss: %f\r", localItems,
View Full Code Here
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
      // randomly initialize our weight matrix by the data we have seen
      DoubleVector feature = tuple.getFirst();
      DoubleVector outcome = tuple.getSecond();
      Iterator<DoubleVectorElement> featureIterator = feature.iterateNonZero();
      while (featureIterator.hasNext()) {
        DoubleVectorElement feat = featureIterator.next();
        Iterator<DoubleVectorElement> outcomeIterator = outcome
            .iterateNonZero();
        while (outcomeIterator.hasNext()) {
          DoubleVectorElement out = outcomeIterator.next();
          theta.set(feat.getIndex(), out.getIndex(), random.nextDouble());
        }
      }
    }
  }
View Full Code Here
105
106
107
108
109
110
111
112
113
114
115
116
117
118
      // probability later on when predicting, here we save a lot space.
      Iterator<DoubleVectorElement> iterateNonZero = rowVector.iterateNonZero();
      double normalizer = FastMath.log(tokenPerClass[row]
          + probabilityMatrix.getColumnCount() - 1);
      while (iterateNonZero.hasNext()) {
        DoubleVectorElement next = iterateNonZero.next();
        double currentWordCount = next.getValue();
        double logProbability = FastMath.log(currentWordCount) - normalizer;
        probabilityMatrix.set(row, next.getIndex(), logProbability);
      }
      if (verbose) {
        System.out
            .println("Computed " + row + " / " + numDistinctClasses + "!");
      }
View Full Code Here
135
136
137
138
139
140
141
142
143
144
145
146
    tokenPerClass[predictedClass] += document.getLength();
    numDocumentsPerClass[predictedClass]++;

    Iterator<DoubleVectorElement> iterateNonZero = document.iterateNonZero();
    while (iterateNonZero.hasNext()) {
      DoubleVectorElement next = iterateNonZero.next();
      double currentCount = probabilityMatrix.get(predictedClass,
          next.getIndex());
      probabilityMatrix.set(predictedClass, next.getIndex(), currentCount
          + next.getValue());
    }
  }
View Full Code Here
152
153
154
155
156
157
158
159
160
161
162
163
164
165
  private double getProbabilityForClass(DoubleVector document, int classIndex) {
    double probabilitySum = 0.0d;
    Iterator<DoubleVectorElement> iterateNonZero = document.iterateNonZero();
    while (iterateNonZero.hasNext()) {
      DoubleVectorElement next = iterateNonZero.next();
      double wordCount = next.getValue();
      double probabilityOfToken = probabilityMatrix.get(classIndex,
          next.getIndex());
      if (probabilityOfToken == 0d) {
        probabilityOfToken = LOW_PROBABILITY;
      }
      probabilitySum += (wordCount * probabilityOfToken);
    }
View Full Code Here
147
148
149
150
151
152
153
154
155
156
157
158
    Arrays.fill(minHashes, Integer.MAX_VALUE);

    for (int i = 0; i < numHashes; i++) {
      Iterator<DoubleVectorElement> iterateNonZero = vector.iterateNonZero();
      while (iterateNonZero.hasNext()) {
        DoubleVectorElement next = iterateNonZero.next();
        int value = (int) next.getValue();
        bytesToHash[0] = (byte) (value >> 24);
        bytesToHash[1] = (byte) (value >> 16);
        bytesToHash[2] = (byte) (value >> 8);
        bytesToHash[3] = (byte) value;
        int hash = functions[i].hash(bytesToHash);
View Full Code Here
22
23
24
25
26
27
28
29
30
31
  @Override
  public Set<Integer> mapDocument(DoubleVector v) {
    Set<Integer> set = new HashSet<>(v.getLength());
    Iterator<DoubleVectorElement> iterateNonZero = v.iterateNonZero();
    while (iterateNonZero.hasNext()) {
      DoubleVectorElement next = iterateNonZero.next();
      set.add(next.getIndex());
    }
    return set;
  }
View Full Code Here
66
67
68
69
70
71
72
73
74
75
76
77
78
      DoubleVector rowVector = transitionProbabilities.getRowVector(rowIndex);
      double sum = rowVector.sum();
      Iterator<DoubleVectorElement> iterateNonZero = rowVector.iterateNonZero();
      // loop over all counts and take the log of the probability
      while (iterateNonZero.hasNext()) {
        DoubleVectorElement columnElement = iterateNonZero.next();
        int columnIndex = columnElement.getIndex();
        double probability = FastMath.log(columnElement.getValue())
            - FastMath.log(sum);
        transitionProbabilities.set(rowIndex, columnIndex, probability);
      }
    }
View Full Code Here
TOP

Related Classes of de.jungblut.math.DoubleVector.DoubleVectorElement

  • de.jungblut.classification.bayes.MultinomialNaiveBayes
  • de.jungblut.classification.regression.LogisticRegression
  • de.jungblut.classification.regression.SparseMultiLabelRegression
  • de.jungblut.datastructure.KDTree
  • de.jungblut.math.activation.AbstractActivationFunction
  • de.jungblut.math.activation.SoftMaxActivationFunction
  • de.jungblut.math.squashing.HammingLossFunction
  • de.jungblut.math.ViterbiUtils
  • de.jungblut.ner.ConditionalLikelihoodCostFunction
  • de.jungblut.nlp.HMM
    • 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.