private void Train(IList <Pair <string, IFileFilter> > trainTreebankPath, Pair <string, IFileFilter> devTreebankPath, string serializedPath)
{
log.Info("Training method: " + op.TrainOptions().trainingMethod);
IList <Tree> binarizedTrees = Generics.NewArrayList();
foreach (Pair <string, IFileFilter> treebank in trainTreebankPath)
{
Sharpen.Collections.AddAll(binarizedTrees, ReadBinarizedTreebank(treebank.First(), treebank.Second()));
}
int nThreads = op.trainOptions.trainingThreads;
nThreads = nThreads <= 0 ? Runtime.GetRuntime().AvailableProcessors() : nThreads;
Edu.Stanford.Nlp.Tagger.Common.Tagger tagger = null;
if (op.testOptions.preTag)
{
Timing retagTimer = new Timing();
tagger = Edu.Stanford.Nlp.Tagger.Common.Tagger.LoadModel(op.testOptions.taggerSerializedFile);
RedoTags(binarizedTrees, tagger, nThreads);
retagTimer.Done("Retagging");
}
ICollection <string> knownStates = FindKnownStates(binarizedTrees);
ICollection <string> rootStates = FindRootStates(binarizedTrees);
ICollection <string> rootOnlyStates = FindRootOnlyStates(binarizedTrees, rootStates);
log.Info("Known states: " + knownStates);
log.Info("States which occur at the root: " + rootStates);
log.Info("States which only occur at the root: " + rootStates);
Timing transitionTimer = new Timing();
IList <IList <ITransition> > transitionLists = CreateTransitionSequence.CreateTransitionSequences(binarizedTrees, op.compoundUnaries, rootStates, rootOnlyStates);
IIndex <ITransition> transitionIndex = new HashIndex <ITransition>();
foreach (IList <ITransition> transitions in transitionLists)
{
transitionIndex.AddAll(transitions);
}
transitionTimer.Done("Converting trees into transition lists");
log.Info("Number of transitions: " + transitionIndex.Size());
Random random = new Random(op.trainOptions.randomSeed);
Treebank devTreebank = null;
if (devTreebankPath != null)
{
devTreebank = ReadTreebank(devTreebankPath.First(), devTreebankPath.Second());
}
PerceptronModel newModel = new PerceptronModel(this.op, transitionIndex, knownStates, rootStates, rootOnlyStates);
newModel.TrainModel(serializedPath, tagger, random, binarizedTrees, transitionLists, devTreebank, nThreads);
this.model = newModel;
}
// TODO: factor out the retagging?
public static void RedoTags(Tree tree, Edu.Stanford.Nlp.Tagger.Common.Tagger tagger)
{
IList <Word> words = tree.YieldWords();
IList <TaggedWord> tagged = tagger.Apply(words);
IList <ILabel> tags = tree.PreTerminalYield();
if (tags.Count != tagged.Count)
{
throw new AssertionError("Tags are not the same size");
}
for (int i = 0; i < tags.Count; ++i)
{
tags[i].SetValue(tagged[i].Tag());
}
}
/// <summary>
/// Will train the model on the given treebank, using devTreebank as
/// a dev set.
/// </summary>
/// <remarks>
/// Will train the model on the given treebank, using devTreebank as
/// a dev set. If op.retrainAfterCutoff is set, will rerun training
/// after the first time through on a limited set of features.
/// </remarks>
public override void TrainModel(string serializedPath, Edu.Stanford.Nlp.Tagger.Common.Tagger tagger, Random random, IList <Tree> binarizedTrees, IList <IList <ITransition> > transitionLists, Treebank devTreebank, int nThreads)
{
if (op.TrainOptions().retrainAfterCutoff&& op.TrainOptions().featureFrequencyCutoff > 0)
{
string tempName = Sharpen.Runtime.Substring(serializedPath, 0, serializedPath.Length - 7) + "-" + "temp.ser.gz";
TrainModel(tempName, tagger, random, binarizedTrees, transitionLists, devTreebank, nThreads, null);
ShiftReduceParser temp = new ShiftReduceParser(op, this);
temp.SaveModel(tempName);
ICollection <string> features = featureWeights.Keys;
featureWeights = Generics.NewHashMap();
TrainModel(serializedPath, tagger, random, binarizedTrees, transitionLists, devTreebank, nThreads, features);
}
else
{
TrainModel(serializedPath, tagger, random, binarizedTrees, transitionLists, devTreebank, nThreads, null);
}
}
public static void RedoTags(IList <Tree> trees, Edu.Stanford.Nlp.Tagger.Common.Tagger tagger, int nThreads)
{
if (nThreads == 1)
{
foreach (Tree tree in trees)
{
RedoTags(tree, tagger);
}
}
else
{
MulticoreWrapper <Tree, Tree> wrapper = new MulticoreWrapper <Tree, Tree>(nThreads, new ShiftReduceParser.RetagProcessor(tagger));
foreach (Tree tree in trees)
{
wrapper.Put(tree);
}
wrapper.Join();
}
}
private void TrainModel(string serializedPath, Edu.Stanford.Nlp.Tagger.Common.Tagger tagger, Random random, IList <Tree> binarizedTrees, IList <IList <ITransition> > transitionLists, Treebank devTreebank, int nThreads, ICollection <string> allowedFeatures
)
{
double bestScore = 0.0;
int bestIteration = 0;
PriorityQueue <ScoredObject <PerceptronModel> > bestModels = null;
if (op.TrainOptions().averagedModels > 0)
{
bestModels = new PriorityQueue <ScoredObject <PerceptronModel> >(op.TrainOptions().averagedModels + 1, ScoredComparator.AscendingComparator);
}
IList <int> indices = Generics.NewArrayList();
for (int i = 0; i < binarizedTrees.Count; ++i)
{
indices.Add(i);
}
Oracle oracle = null;
if (op.TrainOptions().trainingMethod == ShiftReduceTrainOptions.TrainingMethod.Oracle)
{
oracle = new Oracle(binarizedTrees, op.compoundUnaries, rootStates);
}
IList <PerceptronModel.Update> updates = Generics.NewArrayList();
MulticoreWrapper <int, Pair <int, int> > wrapper = null;
if (nThreads != 1)
{
updates = Java.Util.Collections.SynchronizedList(updates);
wrapper = new MulticoreWrapper <int, Pair <int, int> >(op.trainOptions.trainingThreads, new PerceptronModel.TrainTreeProcessor(this, binarizedTrees, transitionLists, updates, oracle));
}
IntCounter <string> featureFrequencies = null;
if (op.TrainOptions().featureFrequencyCutoff > 1)
{
featureFrequencies = new IntCounter <string>();
}
for (int iteration = 1; iteration <= op.trainOptions.trainingIterations; ++iteration)
{
Timing trainingTimer = new Timing();
int numCorrect = 0;
int numWrong = 0;
Java.Util.Collections.Shuffle(indices, random);
for (int start = 0; start < indices.Count; start += op.trainOptions.batchSize)
{
int end = Math.Min(start + op.trainOptions.batchSize, indices.Count);
Triple <IList <PerceptronModel.Update>, int, int> result = TrainBatch(indices.SubList(start, end), binarizedTrees, transitionLists, updates, oracle, wrapper);
numCorrect += result.second;
numWrong += result.third;
foreach (PerceptronModel.Update update in result.first)
{
foreach (string feature in update.features)
{
if (allowedFeatures != null && !allowedFeatures.Contains(feature))
{
continue;
}
Weight weights = featureWeights[feature];
if (weights == null)
{
weights = new Weight();
featureWeights[feature] = weights;
}
weights.UpdateWeight(update.goldTransition, update.delta);
weights.UpdateWeight(update.predictedTransition, -update.delta);
if (featureFrequencies != null)
{
featureFrequencies.IncrementCount(feature, (update.goldTransition >= 0 && update.predictedTransition >= 0) ? 2 : 1);
}
}
}
updates.Clear();
}
trainingTimer.Done("Iteration " + iteration);
log.Info("While training, got " + numCorrect + " transitions correct and " + numWrong + " transitions wrong");
OutputStats();
double labelF1 = 0.0;
if (devTreebank != null)
{
EvaluateTreebank evaluator = new EvaluateTreebank(op, null, new ShiftReduceParser(op, this), tagger);
evaluator.TestOnTreebank(devTreebank);
labelF1 = evaluator.GetLBScore();
log.Info("Label F1 after " + iteration + " iterations: " + labelF1);
if (labelF1 > bestScore)
{
log.Info("New best dev score (previous best " + bestScore + ")");
bestScore = labelF1;
bestIteration = iteration;
}
else
{
log.Info("Failed to improve for " + (iteration - bestIteration) + " iteration(s) on previous best score of " + bestScore);
if (op.trainOptions.stalledIterationLimit > 0 && (iteration - bestIteration >= op.trainOptions.stalledIterationLimit))
{
log.Info("Failed to improve for too long, stopping training");
break;
}
}
log.Info();
if (bestModels != null)
{
bestModels.Add(new ScoredObject <PerceptronModel>(new PerceptronModel(this), labelF1));
if (bestModels.Count > op.TrainOptions().averagedModels)
{
bestModels.Poll();
}
}
}
if (op.TrainOptions().saveIntermediateModels&& serializedPath != null && op.trainOptions.debugOutputFrequency > 0)
{
string tempName = Sharpen.Runtime.Substring(serializedPath, 0, serializedPath.Length - 7) + "-" + Filename.Format(iteration) + "-" + Nf.Format(labelF1) + ".ser.gz";
ShiftReduceParser temp = new ShiftReduceParser(op, this);
temp.SaveModel(tempName);
}
// TODO: we could save a cutoff version of the model,
// especially if we also get a dev set number for it, but that
// might be overkill
if (iteration % 10 == 0 && op.TrainOptions().decayLearningRate > 0.0)
{
learningRate *= op.TrainOptions().decayLearningRate;
}
}
// end for iterations
if (wrapper != null)
{
wrapper.Join();
}
if (bestModels != null)
{
if (op.TrainOptions().cvAveragedModels&& devTreebank != null)
{
IList <ScoredObject <PerceptronModel> > models = Generics.NewArrayList();
while (bestModels.Count > 0)
{
models.Add(bestModels.Poll());
}
Java.Util.Collections.Reverse(models);
double bestF1 = 0.0;
int bestSize = 0;
for (int i_1 = 1; i_1 <= models.Count; ++i_1)
{
log.Info("Testing with " + i_1 + " models averaged together");
// TODO: this is kind of ugly, would prefer a separate object
AverageScoredModels(models.SubList(0, i_1));
ShiftReduceParser temp = new ShiftReduceParser(op, this);
EvaluateTreebank evaluator = new EvaluateTreebank(temp.GetOp(), null, temp, tagger);
evaluator.TestOnTreebank(devTreebank);
double labelF1 = evaluator.GetLBScore();
log.Info("Label F1 for " + i_1 + " models: " + labelF1);
if (labelF1 > bestF1)
{
bestF1 = labelF1;
bestSize = i_1;
}
}
AverageScoredModels(models.SubList(0, bestSize));
}
else
{
AverageScoredModels(bestModels);
}
}
// TODO: perhaps we should filter the features and then get dev
// set scores. That way we can merge the models which are best
// after filtering.
if (featureFrequencies != null)
{
FilterFeatures(featureFrequencies.KeysAbove(op.TrainOptions().featureFrequencyCutoff));
}
CondenseFeatures();
}
public RetagProcessor(Edu.Stanford.Nlp.Tagger.Common.Tagger tagger)
{
this.tagger = tagger;
}
/// <summary>Train a new model.</summary> /// <remarks> /// Train a new model. This is the method to override for new models /// such that the ShiftReduceParser will fill in the model. Given a /// collection of training trees and some other various information, /// this should train a new model. The model is expected to already /// know about the possible transitions and which states are eligible /// to be root states via the BaseModel constructor. /// </remarks> /// <param name="serializedPath">Where serialized models go. If the appropriate options are set, the method can use this to save intermediate models.</param> /// <param name="tagger">The tagger to use when evaluating devTreebank. TODO: it would make more sense for ShiftReduceParser to retag the trees first</param> /// <param name="random">A random number generator to use for any random numbers. Useful to make sure results can be reproduced.</param> /// <param name="binarizedTrainTrees">The treebank to train from.</param> /// <param name="transitionLists">binarizedTrainTrees converted into lists of transitions that will reproduce the same tree.</param> /// <param name="devTreebank">a set of trees which can be used for dev testing (assuming the user provided a dev treebank)</param> /// <param name="nThreads">how many threads the model can use for training</param> public abstract void TrainModel(string serializedPath, Edu.Stanford.Nlp.Tagger.Common.Tagger tagger, Random random, IList <Tree> binarizedTrainTrees, IList <IList <ITransition> > transitionLists, Treebank devTreebank, int nThreads);