Examples of cc.mallet.types.SparseVector

cc.mallet.types.SparseVector
A vector that allocates memory only for non-zero values. When you create a SparseVector, you pass in a list of indices. These are the only elements of the vector you will be allowed to change. The rest are fixed at 0. The interface to Sparse vector uses the concept of a location, which is an integer in the range 0..numLocations which can be mapped to the index (and value) of a non zero element of the vector. A SparseVector can be sparse or dense depending on whether or not an array if indices is specified at construction time. If the SparseVector is dense, the mapping from location to index is the identity mapping. The type of the value an element in a SparseVector (or FeatureVector) can be double or binary (0.0 or 1.0), depending on whether an array of doubles is specified at contruction time. @author Andrew McCallum mccallum@cs.umass.edu

    {
      // Do the defaults first
      defaultConstraints = new SparseVector [templates.length];
      defaultExpectations = new SparseVector [templates.length];
      for (int tidx = 0; tidx < templates.length; tidx++) {
        SparseVector defaults = templates[tidx].getDefaultWeights();
        defaultConstraints[tidx] = (SparseVector) defaults.cloneMatrixZeroed ();
        defaultExpectations[tidx] = (SparseVector) defaults.cloneMatrixZeroed ();
      }


      // And now the others
      constraints = new SparseVector [templates.length][];
      expectations = new SparseVector [templates.length][];

View Full Code Here

        throw new IllegalArgumentException("Argument is not of the " +
                                           " correct dimensions");
      int idx = 0;
      for (int tidx = 0; tidx < templates.length; tidx++) {
        ACRF.Template tmpl = templates [tidx];
        SparseVector defaults = tmpl.getDefaultWeights ();
        double[] values = defaults.getValues();
        System.arraycopy (values, 0, buf, idx, values.length);
        idx += values.length;
      }


      for (int tidx = 0; tidx < templates.length; tidx++) {

View Full Code Here

      cachedValueStale = cachedGradientStale = true;


      int idx = 0;
      for (int tidx = 0; tidx < templates.length; tidx++) {
        ACRF.Template tmpl = templates [tidx];
        SparseVector defaults = tmpl.getDefaultWeights();
        double[] values = defaults.getValues ();
        System.arraycopy (params, idx, values, 0, values.length);
        idx += values.length;
      }


      for (int tidx = 0; tidx < templates.length; tidx++) {

View Full Code Here

      /* Index into current element of cachedGradient[] array. */
      int gidx = 0;


      // First do gradient wrt defaultWeights
      for (int tidx = 0; tidx < templates.length; tidx++) {
        SparseVector theseWeights = templates[tidx].getDefaultWeights ();
        SparseVector theseConstraints = defaultConstraints [tidx];
        SparseVector theseExpectations = defaultExpectations [tidx];
        for (int j = 0; j < theseWeights.numLocations(); j++) {
          double weight = theseWeights.valueAtLocation (j);
          double constraint = theseConstraints.valueAtLocation (j);
          double expectation = theseExpectations.valueAtLocation (j);
          if (printGradient) {
            System.out.println(" gradient ["+gidx+"] = "+constraint+" (ctr) - "+expectation+" (exp) - "+
                             (weight / gaussianPriorVariance)+" (reg)  [feature=DEFAULT]");
          }
          grad [gidx++] = constraint - expectation - (weight / gaussianPriorVariance);
        }
      }


      // Now do other weights
      for (int tidx = 0; tidx < templates.length; tidx++) {
        ACRF.Template tmpl = templates [tidx];
        SparseVector[] weights = tmpl.getWeights ();
        for (int i = 0; i < weights.length; i++) {
          SparseVector thisWeightVec = weights [i];
          SparseVector thisConstraintVec = constraints [tidx][i];
          SparseVector thisExpectationVec = expectations [tidx][i];


          for (int j = 0; j < thisWeightVec.numLocations(); j++) {
            double w = thisWeightVec.valueAtLocation (j);
            double gradient;  // Computed below


            double constraint = thisConstraintVec.valueAtLocation(j);
            double expectation = thisExpectationVec.valueAtLocation(j);


            /* A parameter may be set to -infinity by an external user.
             * We set gradient to 0 because the parameter's value can
             * never change anyway and it will mess up future calculations
             * on the matrix. */

View Full Code Here

    }
  }
  
  public void testPlusEquals ()
  {
    SparseVector s = (SparseVector) s1.cloneMatrix ();
    s.plusEqualsSparse (s2, 2.0);
    checkAnswer (s, new double[] { 3, 5, 7, 6, 7 }); 


    SparseVector s2p = new SparseVector
                       (new int[] { 13 },
                        new double[] { 0.8 });
    s.plusEqualsSparse (s2p, 1.0);
    checkAnswer (s, new double[] { 3, 5, 7, 6.8, 7 }); 


    SparseVector s3p = new SparseVector
                       (new int[] { 14 },
                        new double[] { 0.8 });
    s.plusEqualsSparse (s3p, 1.0);
    checkAnswer (s, new double[] { 3, 5, 7, 6.8, 7 });     // verify s unchanged


    SparseVector s4 = new SparseVector
                      (new int[] { 7, 14, 15 },
                       new double[] { 0.2, 0.8, 1.2 });
    s.plusEqualsSparse (s4, 1.0);
    checkAnswer (s, new double[] { 3, 5, 7.2, 6.8, 8.2 });  


    SparseVector s5 = new SparseVector (new int[] { 7 }, new double[] { 0.2 });
    s5.plusEqualsSparse (s1);
    for (int i = 0; i < s5.numLocations(); i++) {
      assertEquals (7, s5.indexAtLocation (i));
      assertEquals (3.2, s5.valueAtLocation (i), 0.0);
    }


    SparseVector s6 = new SparseVector (new int[] { 7 }, new double[] { 0.2 });
    s6.plusEqualsSparse (s1, 3.5);
    for (int i = 0; i < s6.numLocations(); i++) {
      assertEquals (7, s6.indexAtLocation (i));
      assertEquals (10.7, s6.valueAtLocation (i), 0.0);
    }
  }

View Full Code Here

    // Now add -infinite weight onto a transition, and make sure that it's
    // honored.
    CRF.State state = crf.getState("notstart");
    int widx = crf.getWeightsIndex("BadBad");
    int numFeatures = crf.getInputAlphabet().size();
    SparseVector w = new SparseVector(new double[numFeatures]);
    w.setAll(Double.NEGATIVE_INFINITY);
    crf.setWeights(widx, w);


    state.addWeight(0, "BadBad");
    state.addWeight(1, "BadBad");

View Full Code Here

      assertEquals (10.7, s6.valueAtLocation (i), 0.0);
    }
  }


  public void testDotProduct () {
    SparseVector t1 = new SparseVector (new int[] { 7 }, new double[] { 0.2 });
    assertEquals (0.6, t1.dotProduct (s1), 0.00001);
    assertEquals (0.6, s1.dotProduct (t1), 0.00001);
    
    assertEquals (19.0, s1.dotProduct (s2), 0.00001);
    assertEquals (19.0, s2.dotProduct (s1), 0.00001);


    assertEquals (11.9, s1.dotProduct (d1), 0.00001);
    assertEquals (10.1, s2.dotProduct (d1), 0.00001);


    // test dotproduct when vector with more locations has a lower
    //   max-index than short vector
    SparseVector t2 = new SparseVector (new int[] { 3, 30 }, new double[] { 0.2, 3.5 });
    SparseVector t3 = new SparseVector (null, new double[] { 1, 1, 1, 1, });
    assertEquals (0.2, t3.dotProduct (t2), 0.00001); 
  }

View Full Code Here

    assertEquals (0.2, t3.dotProduct (t2), 0.00001); 
  }


  public void testIncrementValue ()
  {
    SparseVector s = (SparseVector) s1.cloneMatrix ();
    s.incrementValue (5, 0.75);


    double[] ans = new double[] {1, 2.75, 3, 4, 5};
    for (int i = 0; i < s.numLocations(); i++) {
      assertTrue (s.valueAtLocation (i) == ans[i]);
    }
  }

View Full Code Here

  }


  
  public void testSetValue ()
  {
    SparseVector s = (SparseVector) s1.cloneMatrix ();
    s.setValue (5, 0.3);


    double[] ans = new double[] {1, 0.3, 3, 4, 5};
    for (int i = 0; i < s.numLocations(); i++) {
      assertTrue (s.valueAtLocation (i) == ans[i]);
    }
  }

View Full Code Here

    }
  }


  public void testDenseSparseVector ()
  {
    SparseVector svDense = new SparseVector (null, dbl3);
    double sdot = svDense.dotProduct (svDense);
    double ddot = d1.dotProduct (d1);
    assertEquals (sdot, ddot, 0.0001);


    svDense.plusEqualsSparse (s1);
    checkAnswer (svDense, new double[] { 2.0, 2.5, 3.0, 5.7, 3.5,
                                         5.6, 0,   3,   0,   0,
                                         0,   0,   0,   4,   0,
                                         5, });


    svDense.plusEqualsSparse (s1, 2.0);
    checkAnswer (svDense, new double[] { 2.0, 2.5, 3.0, 7.7, 3.5,
                                         9.6, 0,   9,   0,   0,
                                         0,   0,   0,   12,   0,
                                         15, });
    
    double[] dbl4 = new double [dbl3.length + 1];
    for (int i = 0; i < dbl4.length; i++) dbl4[i] = 2.0;
    SparseVector sv4 = new SparseVector (null, dbl4);
    svDense.plusEqualsSparse (sv4);
    checkAnswer (svDense, new double[] { 4.0,  4.5,    5.0,  9.7,   5.5,
                                         11.6, 2.0,   11.0,  2.0,   2.0,
                                         2,   2,   2,   14,   2.0,
                                         17, });

View Full Code Here

0 1 2

TOP

Related Classes of cc.mallet.types.SparseVector

cc.mallet.cluster.KMeans

cc.mallet.fst.CRF

cc.mallet.fst.CRF$TransitionIterator

cc.mallet.fst.tests.TestCRF

cc.mallet.grmm.learning.PiecewiseACRFTrainer$Maxable

cc.mallet.grmm.learning.PseudolikelihoodACRFTrainer$Maxable

cc.mallet.types.tests.TestAugmentableFeatureVector

cc.mallet.types.tests.TestMatrix

cc.mallet.types.tests.TestSparseVector

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.