import cc.mallet.fst.MEMM.State; //导入依赖的package包/类
public void printInstanceLists ()
{
	for (int i = 0; i < memm.numStates(); i++) {
		State state = (State) memm.getState (i);
		InstanceList training = state.trainingSet;
		System.out.println ("State "+i+" : "+state.getName());
		if (training == null) {
			System.out.println ("No data");
			continue;
		}
		for (int j = 0; j < training.size(); j++) {
			Instance inst = training.get (j);
			System.out.println ("From : "+state.getName()+" To : "+inst.getTarget());
			System.out.println ("Instance "+j);
			System.out.println (inst.getTarget());
			System.out.println (inst.getData());
		}
	}
}
 

import cc.mallet.fst.MEMM.State; //导入依赖的package包/类
protected double gatherExpectationsOrConstraints (boolean gatherConstraints)
		{
			// Instance values must either always or never be included in
			// the total values; we can't just sometimes skip a value
			// because it is infinite, this throws off the total values.
			boolean initializingInfiniteValues = false;
			CRF.Factors factors = gatherConstraints ? constraints : expectations;
			CRF.Factors.Incrementor factorIncrementor = factors.new Incrementor ();

			if (infiniteValues == null) {
				infiniteValues = new BitSet ();
				initializingInfiniteValues = true;
			}

			double labelLogProb = 0;
			for (int i = 0; i < memm.numStates(); i++) {
				MEMM.State s = (State) memm.getState (i);

				if (s.trainingSet == null) {
					System.out.println ("Empty training set for state "+s.name);
					continue;
				}

				for (int j = 0; j < s.trainingSet.size(); j++) {
					Instance instance = s.trainingSet.get (j);
					double instWeight = s.trainingSet.getInstanceWeight (j);
					FeatureVector fv = (FeatureVector) instance.getData ();
					String labelString = (String) instance.getTarget ();
					TransitionIterator iter = new TransitionIterator (s, fv, gatherConstraints?labelString:null, memm);
					while (iter.hasNext ()) {
					  // gsc
					  iter.nextState(); // advance the iterator
//						State destination = (MEMM.State) iter.nextState();  // Just to advance the iterator
						double weight = iter.getWeight();
						factorIncrementor.incrementTransition(iter, Math.exp(weight) * instWeight);
						//iter.incrementCount (Math.exp(weight) * instWeight);
						if (!gatherConstraints && iter.getOutput() == labelString) {
							if (!Double.isInfinite (weight))
								labelLogProb += instWeight * weight; // xxx   ?????
							else {
								logger.warning ("State "+i+" transition "+j+" has infinite cost; skipping.");
								if (initializingInfiniteValues)
									throw new IllegalStateException ("Infinite-cost transitions not yet supported"); //infiniteValues.set (j);
								else if (!infiniteValues.get(j))
									throw new IllegalStateException ("Instance i used to have non-infinite value, "
											+"but now it has infinite value.");
							}
						}
					}
				}
			}

			// Force initial & final weight parameters to 0 by making sure that 
			// whether factor refers to expectation or constraint, they have the same value.
			for (int i = 0; i < memm.numStates(); i++) {
				factors.initialWeights[i] = 0.0;
				factors.finalWeights[i] = 0.0;
			}

			return labelLogProb;
		}
 

import cc.mallet.fst.MEMM.State; //导入依赖的package包/类
protected double gatherExpectationsOrConstraints (boolean gatherConstraints)
		{
			// Instance values must either always or never be included in
			// the total values; we can't just sometimes skip a value
			// because it is infinite, this throws off the total values.
			boolean initializingInfiniteValues = false;
			CRF.Factors factors = gatherConstraints ? constraints : expectations;
			CRF.Factors.Incrementor factorIncrementor = factors.new Incrementor ();

			if (infiniteValues == null) {
				infiniteValues = new BitSet ();
				initializingInfiniteValues = true;
			}

			double labelLogProb = 0;
			for (int i = 0; i < memm.numStates(); i++) {
				State s = (State) memm.getState (i);

				if (s.trainingSet == null) {
					System.out.println ("Empty training set for state "+s.name);
					continue;
				}

				for (int j = 0; j < s.trainingSet.size(); j++) {
					Instance instance = s.trainingSet.get (j);
					double instWeight = s.trainingSet.getInstanceWeight (j);
					FeatureVector fv = (FeatureVector) instance.getData ();
					String labelString = (String) instance.getTarget ();
					TransitionIterator iter = new TransitionIterator (s, fv, gatherConstraints?labelString:null, memm);
					while (iter.hasNext ()) {
					  // gsc
					  iter.nextState(); // advance the iterator
//						State destination = (MEMM.State) iter.nextState();  // Just to advance the iterator
						double weight = iter.getWeight();
						factorIncrementor.incrementTransition(iter, Math.exp(weight) * instWeight);
						//iter.incrementCount (Math.exp(weight) * instWeight);
						if (!gatherConstraints && iter.getOutput() == labelString) {
							if (!Double.isInfinite (weight))
								labelLogProb += instWeight * weight; // xxx   ?????
							else {
								logger.warning ("State "+i+" transition "+j+" has infinite cost; skipping.");
								if (initializingInfiniteValues)
									throw new IllegalStateException ("Infinite-cost transitions not yet supported"); //infiniteValues.set (j);
								else if (!infiniteValues.get(j))
									throw new IllegalStateException ("Instance i used to have non-infinite value, "
											+"but now it has infinite value.");
							}
						}
					}
				}
			}

			// Force initial & final weight parameters to 0 by making sure that 
			// whether factor refers to expectation or constraint, they have the same value.
			for (int i = 0; i < memm.numStates(); i++) {
				factors.initialWeights[i] = 0.0;
				factors.finalWeights[i] = 0.0;
			}

			return labelLogProb;
		}