Package cc.mallet.optimize

Examples of cc.mallet.optimize.LimitedMemoryBFGS

119
120
121
122
123
124
125
126
  }
 
  public Optimizer getOptimizer (InstanceList trainingSet) {
    getOptimizableCRF(trainingSet); // this will set this.mcrf if necessary
    if (opt == null || ocrf != opt.getOptimizable())
      opt = new LimitedMemoryBFGS(ocrf);  // Alternative: opt = new ConjugateGradient (0.001);
    return opt;
  }
View Full Code Here
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
    for (; iter < max; iter++) {
      long startTime = System.currentTimeMillis();
     
      // train q
      ConstraintsOptimizableByPR opt = new ConstraintsOptimizableByPR(crf, train, model, numThreads);
      bfgs = new LimitedMemoryBFGS(opt);
      try {
        bfgs.optimize(maxIterPerStep);
      } catch (Exception e) {
        e.printStackTrace();
      }
      opt.shutdown();
     
      /*
      for (int j = 0; j < constraints.size(); j++) {
        constraints.get(j).print();
      }
      */
     
      qValue = opt.getCompleteValueContribution();
      assert(qValue > 0);
     
      // use to train p
      CRFOptimizableByKL optP = new CRFOptimizableByKL(crf, train, model, opt.getCachedDots(), numThreads, 1);
      optP.setGaussianPriorVariance(pGpv);
      LimitedMemoryBFGS bfgsP = new LimitedMemoryBFGS(optP);
     
      try {
        bfgsP.optimize(maxIterPerStep);
      } catch (Exception e) {
        e.printStackTrace();
      }
      optP.shutdown();
     
View Full Code Here
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
    ge.setGaussianPriorVariance(Double.POSITIVE_INFINITY);

    CRFOptimizableByGradientValues opt =
      new CRFOptimizableByGradientValues(crf,new Optimizable.ByGradientValue[] { optLikelihood, ge });
   
    LimitedMemoryBFGS optimizer = new LimitedMemoryBFGS(opt);

    try {
      converged = optimizer.optimize(numIterations);
    }
    catch (Exception e) {
      e.printStackTrace();
    }
   
    optimizer.reset();
    try {
      converged = optimizer.optimize(numIterations);
    }
    catch (Exception e) {
      e.printStackTrace();
    }
   
View Full Code Here
115
116
117
118
119
120
121
122
123
    return optimizable;
  }

  public Optimizer getOptimizer(Optimizable.ByGradientValue optimizable) {
    if (optimizer == null) {
      optimizer = new LimitedMemoryBFGS(optimizable);
    }
    return optimizer;
  }
View Full Code Here
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
    if (iteration == 0) {
      // train with log-likelihood only first
      CRFOptimizableByLabelLikelihood likelihood =
        new CRFOptimizableByLabelLikelihood(crf, labeled);
      likelihood.setGaussianPriorVariance(gaussianPriorVariance);
      this.bfgs = new LimitedMemoryBFGS(likelihood);
      logger.info ("CRF about to train with "+numIterations+" iterations");
      for (int i = 0; i < numIterations; i++) {
        try {
          converged = bfgs.optimize(1);
          iteration++;
          logger.info ("CRF finished one iteration of maximizer, i="+i);
          runEvaluators();
        } catch (IllegalArgumentException e) {
          e.printStackTrace();
          logger.info ("Catching exception; saying converged.");
          converged = true;
        } catch (Exception e) {
          e.printStackTrace();
          logger.info("Catching exception; saying converged.");
          converged = true;
        }
        if (converged) {
          logger.info ("CRF training has converged, i="+i);
          break;
        }
      }
      iteration = 0;
    }
   
    // train with log-likelihood + entropy regularization
    CRFOptimizableByLabelLikelihood likelihood = new CRFOptimizableByLabelLikelihood(crf, labeled);
    likelihood.setGaussianPriorVariance(gaussianPriorVariance);
    CRFOptimizableByEntropyRegularization regularization = new CRFOptimizableByEntropyRegularization(crf, unlabeled);
    regularization.setScalingFactor(this.entRegScalingFactor);
   
    CRFOptimizableByGradientValues regLikelihood = new CRFOptimizableByGradientValues(crf,
        new Optimizable.ByGradientValue[] { likelihood, regularization} );
    this.bfgs = new LimitedMemoryBFGS(regLikelihood);
    converged = false;
    logger.info ("CRF about to train with "+numIterations+" iterations");
    // sometimes resetting the optimizer helps to find
    // a better parameter setting
    for (int reset = 0; reset < DEFAULT_NUM_RESETS + 1; reset++) {
View Full Code Here
208
209
210
211
212
213
214
215
216
217
218
219
220
221
   
    // train supervised
    MaxEntOptimizableByLabelDistribution opt = new  MaxEntOptimizableByLabelDistribution(dataLabeled,p);
    opt.setGaussianPriorVariance(pGPV);

    LimitedMemoryBFGS bfgs = new LimitedMemoryBFGS(opt);
    try { bfgs.optimize(); } catch (Exception e) { e.printStackTrace(); }
    bfgs.reset();
    try { bfgs.optimize(); } catch (Exception e) { e.printStackTrace(); }
   
    double value = 0;
    for (MaxEntPRConstraint constraint : q.getConstraintFeatures()) {
      // plus sign because this returns negative values
      value += constraint.getCompleteValueContribution();
View Full Code Here
238
239
240
241
242
243
244
245
246
247
248
249
      }
    }
   
    PRAuxClassifierOptimizable optimizable = new PRAuxClassifierOptimizable(data,base,q);
   
    LimitedMemoryBFGS bfgs = new LimitedMemoryBFGS(optimizable);
    try { bfgs.optimize(); } catch (Exception e) { e.printStackTrace(); }
    bfgs.reset();
    try { bfgs.optimize(); } catch (Exception e) { e.printStackTrace(); }
   
    return base;
  }
View Full Code Here
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
        DMROptimizable optimizable = new DMROptimizable(parameterInstances, dmrParameters);
        optimizable.setRegularGaussianPriorVariance(0.5);
        optimizable.setInterceptGaussianPriorVariance(100.0);

    LimitedMemoryBFGS optimizer = new LimitedMemoryBFGS(optimizable);

    // Optimize once
    try {
      optimizer.optimize();
    } catch (IllegalArgumentException e) {
      // step size too small
    }

    // Restart with a fresh initialization to improve likelihood
    try {
      optimizer.optimize();
    } catch (IllegalArgumentException e) {
      // step size too small
    }
        dmrParameters = optimizable.getClassifier();
View Full Code Here
103
104
105
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
    if (iteration == 0) {
      // train with log-likelihood only first
      CRFOptimizableByLabelLikelihood likelihood =
        new CRFOptimizableByLabelLikelihood(crf, labeled);
      likelihood.setGaussianPriorVariance(gaussianPriorVariance);
      this.bfgs = new LimitedMemoryBFGS(likelihood);
      logger.info ("CRF about to train with "+numIterations+" iterations");
      for (int i = 0; i < numIterations; i++) {
        try {
          converged = bfgs.optimize(1);
          iteration++;
          logger.info ("CRF finished one iteration of maximizer, i="+i);
          runEvaluators();
        } catch (IllegalArgumentException e) {
          e.printStackTrace();
          logger.info ("Catching exception; saying converged.");
          converged = true;
        } catch (Exception e) {
          e.printStackTrace();
          logger.info("Catching exception; saying converged.");
          converged = true;
        }
        if (converged) {
          logger.info ("CRF training has converged, i="+i);
          break;
        }
      }
      iteration = 0;
    }
   
    // train with log-likelihood + entropy regularization
    CRFOptimizableByLabelLikelihood likelihood = new CRFOptimizableByLabelLikelihood(crf, labeled);
    likelihood.setGaussianPriorVariance(gaussianPriorVariance);
    CRFOptimizableByEntropyRegularization regularization = new CRFOptimizableByEntropyRegularization(crf, unlabeled);
    regularization.setScalingFactor(this.entRegScalingFactor);
   
    CRFOptimizableByGradientValues regLikelihood = new CRFOptimizableByGradientValues(crf,
        new Optimizable.ByGradientValue[] { likelihood, regularization} );
    this.bfgs = new LimitedMemoryBFGS(regLikelihood);
    converged = false;
    logger.info ("CRF about to train with "+numIterations+" iterations");
    // sometimes resetting the optimizer helps to find
    // a better parameter setting
    for (int reset = 0; reset < DEFAULT_NUM_RESETS + 1; reset++) {
View Full Code Here
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
   
    CRFOptimizableByGECriteria ge =
      new CRFOptimizableByGECriteria(criteria, crf, unlabeledSet, numThreads);
    ge.setGaussianPriorVariance(gaussianPriorVariance);
   
    LimitedMemoryBFGS bfgs = new LimitedMemoryBFGS(ge);
   
    converged = false;
    logger.info ("CRF about to train with "+numIterations+" iterations");
    // sometimes resetting the optimizer helps to find
    // a better parameter setting
    int iter = 0;
    for (int reset = 0; reset < DEFAULT_NUM_RESETS + 1; reset++) {
      for (; iter < numIterations; iter++) {
        try {
          converged = bfgs.optimize (1);
          iteration++;
          logger.info ("CRF finished one iteration of maximizer, i="+iter);
          runEvaluators();
        } catch (IllegalArgumentException e) {
          e.printStackTrace();
          logger.info ("Catching exception; saying converged.");
          converged = true;
        } catch (Exception e) {
          e.printStackTrace();
          logger.info("Catching exception; saying converged.");
          converged = true;
        }
        if (converged) {
          logger.info ("CRF training has converged, i="+iter);
          break;
        }
      }
      bfgs.reset();
    }
   
    ge.shutdown();
   
    return converged;
View Full Code Here
TOP

Related Classes of cc.mallet.optimize.LimitedMemoryBFGS

  • cc.mallet.classify.MaxEntGERangeTrainer
  • cc.mallet.classify.MaxEntGETrainer
  • cc.mallet.classify.MaxEntPRTrainer
  • cc.mallet.classify.MaxEntTrainer
  • cc.mallet.classify.MCMaxEntTrainer
  • cc.mallet.classify.RankMaxEntTrainer
  • cc.mallet.fst.CRFTrainerByLabelLikelihood
  • cc.mallet.fst.CRFTrainerByThreadedLabelLikelihood
  • cc.mallet.fst.CRFTrainerByValueGradients
  • cc.mallet.fst.MEMMTrainer
    • 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.