Package water.fvec

Examples of water.fvec.Vec$CollectDomain

85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
      throw new IllegalArgumentException("vpredict cannot be class labels, expect probabilities.");
  }

  public void execImpl() {
    init();
    Vec va = null, vp;
    try {
      va = vactual.toEnum(); // always returns TransfVec
      vp = vpredict;
      // The vectors are from different groups => align them, but properly delete it after computation
      if (!va.group().equals(vp.group())) {
        vp = va.align(vp);
      }
      // compute thresholds, if not user-given
      if (thresholds != null) {
        sort(thresholds);
        if (ArrayUtils.minValue(thresholds) < 0) throw new IllegalArgumentException("Minimum threshold cannot be negative.");
        if (ArrayUtils.maxValue(thresholds) > 1) throw new IllegalArgumentException("Maximum threshold cannot be greater than 1.");
      } else {
        HashSet hs = new HashSet();
        final int bins = (int)Math.min(vpredict.length(), 200l);
        final long stride = Math.max(vpredict.length() / bins, 1);
        for( int i=0; i<bins; ++i) hs.add(new Float(vpredict.at(i*stride))); //data-driven thresholds TODO: use percentiles (from Summary2?)
        for (int i=0;i<51;++i) hs.add(new Float(i/50.)); //always add 0.02-spaced thresholds from 0 to 1

        // created sorted vector of unique thresholds
        thresholds = new float[hs.size()];
        int i=0;
        for (Object h : hs) {thresholds[i++] = (Float)h; }
        sort(thresholds);
      }
      // compute CMs
      aucdata = new AUCData().compute(new AUCTask(thresholds,va.mean()).doAll(va,vp).getCMs(), thresholds, va.factors(), threshold_criterion);
    } finally {       // Delete adaptation vectors
      if (va!=null) DKV.remove(va._key);
    }
  }
View Full Code Here
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
      throw new IllegalArgumentException("Predicted vector cannot be categorical for regression scoring.");
  }

  public void execImpl() {
    init();
    Vec va = null,vp = null, avp = null;
    try {
      if (classification) {
        // Create a new vectors - it is cheap since vector are only adaptation vectors
        va = vactual .toEnum(); // always returns TransfVec
        actual_domain = va.factors();
        vp = vpredict.toEnum(); // always returns TransfVec
        predicted_domain = vp.factors();
        if (!Arrays.equals(actual_domain, predicted_domain)) {
          domain = ArrayUtils.domainUnion(actual_domain, predicted_domain);
          int[][] vamap = Model.getDomainMapping(domain, actual_domain, true);
          va = TransfVec.compose( (TransfVec) va, vamap, domain, false ); // delete original va
          int[][] vpmap = Model.getDomainMapping(domain, predicted_domain, true);
          vp = TransfVec.compose( (TransfVec) vp, vpmap, domain, false ); // delete original vp
        } else domain = actual_domain;
        // The vectors are from different groups => align them, but properly delete it after computation
        if (!va.group().equals(vp.group())) {
          avp = vp;
          vp = va.align(vp);
        }
        cm = new CM(domain.length).doAll(va,vp)._cm;
      } else {
        mse = new CM(1).doAll(vactual,vpredict).mse();
      }
View Full Code Here
41
42
43
44
45
46
47
48
49
50
51
    if( !val.isFrame() ) throw new IllegalArgumentException("Not a Frame");
    Frame fr = val.get();

    // Peel out an optional column; restrict to this column
    if( column != null ) {
      Vec vec = fr.vec(column);
      if( vec==null ) throw new IllegalArgumentException("Column "+column+" not found in frame "+key);
      fr = new Frame(new String[]{column}, new Vec[]{vec});
    }

    f._fr = fr;
View Full Code Here
76
77
78
79
80
81
82
83
84
85
86
  /** Return a single column from the frame. */
  protected Schema column(int version, Frames f) { // TODO: should return a Vec schema
    Frame frame = getFromDKV(f.key);

    // TODO: We really want to return a different schema here!
    Vec vec = frame.vec(f.column);
    if (null == vec)
      throw new IllegalArgumentException("Did not find column: " + f.column + " in frame: " + f.key.toString());

    Vec[] vecs = { vec };
    String[] names = { f.column };
View Full Code Here
90
91
92
93
94
95
96
97
98
99
100
    return this.schema(version).fillFromImpl(f);
  }

  protected FramesBase columnSummary(int version, Frames frames) {
    Frame frame = getFromDKV(frames.key);
    Vec vec = frame.vec(frames.column);
    if (null == vec)
      throw new IllegalArgumentException("Did not find column: " + frames.column + " in frame: " + frames.key.toString());

    // Compute second pass of rollups: the histograms.  Side-effects the Vec.
    // TODO: side effects, ugh.
View Full Code Here
52
53
54
55
56
57
58
59
60
61
62
    _adaptedValidation = av[0];
//    gtrash(av[1]); // delete this after computation
    if (_fromValid2CM!=null) {
      assert classification : "Validation response transformation should be declared only for classification!";
      assert _fromModel2CM != null : "Model response transformation should exist if validation response transformation exists!";
      Vec tmp = _validResponse.toEnum();
      _adaptedValidationResponse = tmp.makeTransf(_fromValid2CM, getCMDomain()); // Add an original response adapted to CM domain
//      gtrash(_adaptedValidationResponse); // Add the created vector to a clean-up list
//      gtrash(tmp);
    }
  }
View Full Code Here
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
   * @param adaptFrm
   * @return
   */
  private Frame scoreImpl(Frame adaptFrm) {
    int ridx = adaptFrm.find(_output.responseName());
    Vec vecs[] = adaptFrm.vecs();
    assert ridx == -1 : "Adapted frame should not contain response in scoring method!";
    assert _output.nfeatures() == adaptFrm.numCols() : "Number of model features " + _output.nfeatures() + " != number of test set columns: " + adaptFrm.numCols();
    assert vecs.length == _output._names.length-1 : "Scoring data set contains wrong number of columns: " + vecs.length  + " instead of " + (_output._names.length-1);

    // Create a new vector for response
    // If the model produces a classification/enum, copy the domain into the
    // result vector.
    Vec v = adaptFrm.anyVec().makeZero(_output.classNames());
    adaptFrm.add("predict",v);
    if( _output.nclasses() > 1 ) {
      String prefix = "";
      for( int c=0; c<_output.nclasses(); c++ ) // if any class is the same as column name in frame, then prefix all classnames
        if (ArrayUtils.contains(adaptFrm._names, _output.classNames()[c])) { prefix = "class_"; break; }
View Full Code Here
314
315
316
317
318
319
320
321
322
323
324
    ArrayList<Vec> avecs = new ArrayList<>(); // adapted vectors
    ArrayList<String> anames = new ArrayList<>(); // names for adapted vector

    for( int c=0; c<map.length; c++ ) // Iterate over columns
      if(map[c] != null) { // Column needs adaptation
        Vec adaptedVec;
        if (toEnum[c]) { // Vector was flipped to column already, compose transformation
          adaptedVec = TransfVec.compose((TransfVec) frvecs[c], map[c], vfr.domains()[c], false);
        } else adaptedVec = frvecs[c].makeTransf(map[c], vfr.domains()[c]);
        avecs.add(frvecs[c] = adaptedVec);
        anames.add(names[c]); // Collect right names
View Full Code Here
59
60
61
62
63
64
65
66
67
68
69
70
71
72
  }

//  @Override
  public void execImpl() {
    init();
    Vec va = null;
    try {
      va = vactual.toEnum(); // always returns TransfVec
      actual_domain = va.factors();
      if (max_k > predict.numCols()-1) {
        Log.warn("Reducing Hitratio Top-K value to maximum value allowed: " + String.format("%,d", predict.numCols() - 1));
        max_k = predict.numCols() - 1;
      }
      final Frame actual_predict = new Frame(predict.names().clone(), predict.vecs().clone());
View Full Code Here
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
   * @param adaptFrm
   * @return
   */
  private Frame scoreImpl(Frame adaptFrm) {
    int ridx = adaptFrm.find(responseName());
    Vec vecs[] = adaptFrm.vecs();
    assert ridx == -1 : "Adapted frame should not contain response in scoring method!";
    assert nfeatures() == adaptFrm.numCols() : "Number of model features " + nfeatures() + " != number of test set columns: " + adaptFrm.numCols();
    assert vecs.length == _names.length-1 : "Scoring data set contains wrong number of columns: " + vecs.length  + " instead of " + (_names.length-1);

    // Create a new vector for response
    // If the model produces a classification/enum, copy the domain into the
    // result vector.
    Vec v = adaptFrm.anyVec().makeZero(classNames());
    adaptFrm.add("predict",v);
    if( nclasses() > 1 ) {
      String prefix = "";
      for( int c=0; c<nclasses(); c++ ) // if any class is the same as column name in frame, then prefix all classnames
        if (ArrayUtils.contains(adaptFrm._names, classNames()[c])) { prefix = "class_"; break; }
View Full Code Here
TOP

Related Classes of water.fvec.Vec$CollectDomain

  • org.apache.mahout.h2obindings.H2OHdfs
  • org.apache.mahout.h2obindings.H2OHelper
  • org.apache.mahout.h2obindings.ops.ABt
  • org.apache.mahout.h2obindings.ops.AewB
  • org.apache.mahout.h2obindings.ops.AewScalar
  • org.apache.mahout.h2obindings.ops.At
  • org.apache.mahout.h2obindings.ops.AtA
  • org.apache.mahout.h2obindings.ops.AtB
  • org.apache.mahout.h2obindings.ops.Ax
  • org.apache.mahout.h2obindings.ops.Cbind
    • 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.