📄 transducer.java
字号:
@param constrainedSequence lattice must have labels of this sequence from <code> requiredSegment.start </code> to <code> requiredSegment.end </code> correctly */ public Lattice forwardBackward (Sequence inputSequence, Sequence outputSequence, Segment requiredSegment, Sequence constrainedSequence) { if (constrainedSequence.size () != inputSequence.size ()) throw new IllegalArgumentException ("constrainedSequence.size [" + constrainedSequence.size () + "] != inputSequence.size [" + inputSequence.size () + "]"); // constraints tells the lattice which states must emit which // observations. positive values say all paths must pass through // this state index, negative values say all paths must _not_ // pass through this state index. 0 means we don't // care. initialize to 0. include 1 extra node for start state. int [] constraints = new int [constrainedSequence.size() + 1]; for (int c = 0; c < constraints.length; c++) constraints[c] = 0; for (int i=requiredSegment.getStart (); i <= requiredSegment.getEnd(); i++) { int si = stateIndexOfString ((String)constrainedSequence.get (i)); if (si == -1) logger.warning ("Could not find state " + constrainedSequence.get (i) + ". Check that state labels match startTages and inTags, and that all labels are seen in training data.");// throw new IllegalArgumentException ("Could not find state " + constrainedSequence.get(i) + ". Check that state labels match startTags and InTags."); constraints[i+1] = si + 1; } // set additional negative constraint to ensure state after // segment is not a continue tag // xxx if segment length=1, this actually constrains the sequence // to B-tag (B-tag)', instead of the intended constraint of B-tag // (I-tag)' // the fix below is unsafe, but will have to do for now. // FIXED BELOW/* String endTag = (String) constrainedSequence.get (requiredSegment.getEnd ()); if (requiredSegment.getEnd()+2 < constraints.length) { if (requiredSegment.getStart() == requiredSegment.getEnd()) { // segment has length 1 if (endTag.startsWith ("B-")) { endTag = "I" + endTag.substring (1, endTag.length()); } else if (!(endTag.startsWith ("I-") || endTag.startsWith ("0"))) throw new IllegalArgumentException ("Constrained Lattice requires that states are tagged in B-I-O format."); } int statei = stateIndexOfString (endTag); if (statei == -1) // no I- tag for this B- tag statei = stateIndexOfString ((String)constrainedSequence.get (requiredSegment.getStart ())); constraints[requiredSegment.getEnd() + 2] = - (statei + 1); }*/ if (requiredSegment.getEnd() + 2 < constraints.length) { // if String endTag = requiredSegment.getInTag().toString(); int statei = stateIndexOfString (endTag); if (statei == -1) throw new IllegalArgumentException ("Could not find state " + endTag + ". Check that state labels match startTags and InTags."); constraints[requiredSegment.getEnd() + 2] = - (statei + 1); } // printStates (); logger.fine ("Segment:\n" + requiredSegment.sequenceToString () + "\nconstrainedSequence:\n" + constrainedSequence + "\nConstraints:\n"); for (int i=0; i < constraints.length; i++) { logger.fine (constraints[i] + "\t"); } logger.fine (""); return forwardBackward (inputSequence, outputSequence, constraints); } public int stateIndexOfString (String s) { for (int i = 0; i < this.numStates(); i++) { String state = this.getState (i).getName(); if (state.equals (s)) return i; } return -1; } private void printStates () { for (int i = 0; i < this.numStates(); i++) logger.fine (i + ":" + this.getState (i).getName()); } public void print () { logger.fine ("Transducer "+this); printStates(); } public Lattice forwardBackward (Sequence inputSequence, Sequence outputSequence, int [] constraints) { return new Lattice (inputSequence, outputSequence, false, null, constraints); } // Remove this method? // If "increment" is true, call incrementInitialCount, incrementFinalCount and incrementCount private Lattice forwardBackward (SequencePair inputOutputPair, boolean increment) { return this.forwardBackward (inputOutputPair.input(), inputOutputPair.output(), increment); } // xxx Include methods like this? // ...making random selections proportional to cost //public Transduction transduce (Object[] inputSequence) //{ throw new UnsupportedOperationException (); } //public Transduction transduce (Sequence inputSequence) //{ throw new UnsupportedOperationException (); } public class Lattice // ?? extends SequencePairAlignment, but there isn't just a single output! { // "ip" == "input position", "op" == "output position", "i" == "state index" double cost; Sequence input, output; LatticeNode[][] nodes; // indexed by ip,i int latticeLength; // xxx Now that we are incrementing here directly, there isn't // necessarily a need to save all these arrays... // -log(probability) of being in state "i" at input position "ip" double[][] gammas; // indexed by ip,i LabelVector labelings[]; // indexed by op, created only if "outputAlphabet" is non-null in constructor private LatticeNode getLatticeNode (int ip, int stateIndex) { if (nodes[ip][stateIndex] == null) nodes[ip][stateIndex] = new LatticeNode (ip, getState (stateIndex)); return nodes[ip][stateIndex]; } // You may pass null for output, meaning that the lattice // is not constrained to match the output protected Lattice (Sequence input, Sequence output, boolean increment) { this (input, output, increment, null); } // If outputAlphabet is non-null, this will create a LabelVector // for each position in the output sequence indicating the // probability distribution over possible outputs at that time // index protected Lattice (Sequence input, Sequence output, boolean increment, LabelAlphabet outputAlphabet) { if (false && logger.isLoggable (Level.FINE)) { logger.fine ("Starting Lattice"); logger.fine ("Input: "); for (int ip = 0; ip < input.size(); ip++) logger.fine (" " + input.get(ip)); logger.fine ("\nOutput: "); if (output == null) logger.fine ("null"); else for (int op = 0; op < output.size(); op++) logger.fine (" " + output.get(op)); logger.fine ("\n"); } // Initialize some structures this.input = input; this.output = output; // xxx Not very efficient when the lattice is actually sparse, // especially when the number of states is large and the // sequence is long. latticeLength = input.size()+1; int numStates = numStates(); nodes = new LatticeNode[latticeLength][numStates]; // xxx Yipes, this could get big; something sparse might be better? gammas = new double[latticeLength][numStates]; // xxx Move this to an ivar, so we can save it? But for what? // Commenting this out, because it's a memory hog and not used right now. // Uncomment and conditionalize under a flag if ever needed. -cas // double xis[][][] = new double[latticeLength][numStates][numStates]; double outputCounts[][] = null; if (outputAlphabet != null) outputCounts = new double[latticeLength][outputAlphabet.size()]; for (int i = 0; i < numStates; i++) { for (int ip = 0; ip < latticeLength; ip++) gammas[ip][i] = INFINITE_COST; /* Commenting out xis -cas for (int j = 0; j < numStates; j++) for (int ip = 0; ip < latticeLength; ip++) xis[ip][i][j] = INFINITE_COST; */ } // Forward pass logger.fine ("Starting Foward pass"); boolean atLeastOneInitialState = false; for (int i = 0; i < numStates; i++) { double initialCost = getState(i).initialCost; //System.out.println ("Forward pass initialCost = "+initialCost); if (initialCost < INFINITE_COST) { getLatticeNode(0, i).alpha = initialCost; //System.out.println ("nodes[0][i].alpha="+nodes[0][i].alpha); atLeastOneInitialState = true; } } if (atLeastOneInitialState == false) logger.warning ("There are no starting states!"); for (int ip = 0; ip < latticeLength-1; ip++) for (int i = 0; i < numStates; i++) { if (nodes[ip][i] == null || nodes[ip][i].alpha == INFINITE_COST) // xxx if we end up doing this a lot, // we could save a list of the non-null ones continue; State s = getState(i); TransitionIterator iter = s.transitionIterator (input, ip, output, ip); if (logger.isLoggable (Level.FINE)) logger.fine (" Starting Foward transition iteration from state " + s.getName() + " on input " + input.get(ip).toString() + " and output " + (output==null ? "(null)" : output.get(ip).toString())); while (iter.hasNext()) { State destination = iter.nextState(); if (logger.isLoggable (Level.FINE)) logger.fine ("Forward Lattice[inputPos="+ip +"][source="+s.getName() +"][dest="+destination.getName()+"]"); LatticeNode destinationNode = getLatticeNode (ip+1, destination.getIndex()); destinationNode.output = iter.getOutput(); double transitionCost = iter.getCost(); if (logger.isLoggable (Level.FINE)) logger.fine ("transitionCost="+transitionCost +" nodes["+ip+"]["+i+"].alpha="+nodes[ip][i].alpha +" destinationNode.alpha="+destinationNode.alpha); destinationNode.alpha = sumNegLogProb (destinationNode.alpha, nodes[ip][i].alpha + transitionCost); //System.out.println ("destinationNode.alpha <- "+destinationNode.alpha); } } // Calculate total cost of Lattice. This is the normalizer cost = INFINITE_COST; for (int i = 0; i < numStates; i++) if (nodes[latticeLength-1][i] != null) { // Note: actually we could sum at any ip index, // the choice of latticeLength-1 is arbitrary //System.out.println ("Ending alpha, state["+i+"] = "+nodes[latticeLength-1][i].alpha); //System.out.println ("Ending beta, state["+i+"] = "+getState(i).finalCost); cost = sumNegLogProb (cost, (nodes[latticeLength-1][i].alpha + getState(i).finalCost)); } // Cost is now an "unnormalized cost" of the entire Lattice //assert (cost >= 0) : "cost = "+cost; // If the sequence has infinite cost, just return. // Usefully this avoids calling any incrementX methods. // It also relies on the fact that the gammas[][] and .alpha and .beta values // are already initialized to values that reflect infinite cost // xxx Although perhaps not all (alphas,betas) exactly correctly reflecting? if (cost == INFINITE_COST) return; // Backward pass for (int i = 0; i < numStates; i++) if (nodes[latticeLength-1][i] != null) { State s = getState(i); nodes[latticeLength-1][i].beta = s.finalCost; gammas[latticeLength-1][i] = nodes[latticeLength-1][i].alpha + nodes[latticeLength-1][i].beta - cost; if (increment) { double p = Math.exp(-gammas[latticeLength-1][i]); assert (p < INFINITE_COST && !Double.isNaN(p)) : "p="+p+" gamma="+gammas[latticeLength-1][i]; s.incrementFinalCount (p); } } for (int ip = latticeLength-2; ip >= 0; ip--) { for (int i = 0; i < numStates; i++) { if (nodes[ip][i] == null || nodes[ip][i].alpha == INFINITE_COST) // Note that skipping here based on alpha means that beta values won't // be correct, but since alpha is infinite anyway, it shouldn't matter. continue; State s = getState(i); TransitionIterator iter = s.transitionIterator (input, ip, output, ip); while (iter.hasNext()) { State destination = iter.nextState(); if (logger.isLoggable (Level.FINE)) logger.fine ("Backward Lattice[inputPos="+ip +"][source="+s.getName() +"][dest="+destination.getName()+"]"); int j = destination.getIndex(); LatticeNode destinationNode = nodes[ip+1][j]; if (destinationNode != null) { double transitionCost = iter.getCost(); assert (!Double.isNaN(transitionCost)); // assert (transitionCost >= 0); Not necessarily double oldBeta = nodes[ip][i].beta; assert (!Double.isNaN(nodes[ip][i].beta)); nodes[ip][i].beta = sumNegLogProb (nodes[ip][i].beta, destinationNode.beta + transitionCost); assert (!Double.isNaN(nodes[ip][i].beta)) : "dest.beta="+destinationNode.beta+" trans="+transitionCost+" sum="+(destinationNode.beta+transitionCost) + " oldBeta="+oldBeta;// xis[ip][i][j] = nodes[ip][i].alpha + transitionCost + nodes[ip+1][j].beta - cost; assert (!Double.isNaN(nodes[ip][i].alpha)); assert (!Double.isNaN(transitionCost)); assert (!Double.isNaN(nodes[ip+1][j].beta)); assert (!Double.isNaN(cost)); if (increment || outputAlphabet != null) { double xi = nodes[ip][i].alpha + transitionCost + nodes[ip+1][j].beta - cost; double p = Math.exp(-xi); assert (p < INFINITE_COST && !Double.isNaN(p)) : "xis["+ip+"]["+i+"]["+j+"]="+-xi; if (increment) iter.incrementCount (p); if (outputAlphabet != null) { int outputIndex = outputAlphabet.lookupIndex (iter.getOutput(), false); assert (outputIndex >= 0); // xxx This assumes that "ip" == "op"! outputCounts[ip][outputIndex] += p; //System.out.println ("CRF Lattice outputCounts["+ip+"]["+outputIndex+"]+="+p); } } } } gammas[ip][i] = nodes[ip][i].alpha + nodes[ip][i].beta - cost; } } if (increment) for (int i = 0; i < numStates; i++) { double p = Math.exp(-gammas[0][i]); assert (p < INFINITE_COST && !Double.isNaN(p)); getState(i).incrementInitialCount (p); } if (outputAlphabet != null) { labelings = new LabelVector[latticeLength]; for (int ip = latticeLength-2; ip >= 0; ip--) { assert (Math.abs(1.0-DenseVector.sum (outputCounts[ip])) < 0.000001);; labelings[ip] = new LabelVector (outputAlphabet, outputCounts[ip]); } } } // culotta: constructor for constrained lattice /** Create a lattice that constrains its transitions such that the * <position,label> pairs in "constraints" are adhered * to. constraints is an array where each entry is the index of * the required label at that position. An entry of 0 means there * are no constraints on that <position, label>. Positive values * mean the path must pass through that state. Negative values * mean the path must _not_ pass through that state. NOTE - * constraints.length must be equal to output.size() + 1. A * lattice has one extra position for the initial * state. Generally, this should be unconstrained, since it does * not produce an observation. */ protected Lattice (Sequence input, Sequence output, boolean increment, LabelAlphabet outputAlphabet, int [] constraints)
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -