edu.stanford.nlp.parser.shiftreduce

Class BaseModel

java.lang.Object

edu.stanford.nlp.parser.shiftreduce.BaseModel

All Implemented Interfaces:

Serializable

Direct Known Subclasses:

PerceptronModel

public abstract class BaseModel
extends Object
implements Serializable

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
BaseModel(BaseModel other)
BaseModel(ShiftReduceOptions op, Index<Transition> transitionIndex, Set<String> knownStates, Set<String> rootStates, Set<String> rootOnlyStates)

Method Summary

Modifier and Type	Method and Description
Transition	findEmergencyTransition(State state, List<ParserConstraint> constraints) Returns a transition which might not even be part of the model, but will hopefully allow progress in an otherwise stuck parse TODO: perhaps we want to create an EmergencyTransition class which indicates that something has gone wrong
abstract Collection<ScoredObject<Integer>>	findHighestScoringTransitions(State state, boolean requireLegal, int numTransitions, List<ParserConstraint> constraints)
abstract void	trainModel(String serializedPath, Tagger tagger, Random random, List<Tree> binarizedTrainTrees, List<List<Transition>> transitionLists, Treebank devTreebank, int nThreads) Train a new model.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

BaseModel

public BaseModel(ShiftReduceOptions op,
                 Index<Transition> transitionIndex,
                 Set<String> knownStates,
                 Set<String> rootStates,
                 Set<String> rootOnlyStates)

BaseModel

public BaseModel(BaseModel other)

Method Detail

findEmergencyTransition

public Transition findEmergencyTransition(State state,
                                          List<ParserConstraint> constraints)

Returns a transition which might not even be part of the model, but will hopefully allow progress in an otherwise stuck parse TODO: perhaps we want to create an EmergencyTransition class which indicates that something has gone wrong

findHighestScoringTransitions

public abstract Collection<ScoredObject<Integer>> findHighestScoringTransitions(State state,
                                                                                boolean requireLegal,
                                                                                int numTransitions,
                                                                                List<ParserConstraint> constraints)

trainModel

public abstract void trainModel(String serializedPath,
                                Tagger tagger,
                                Random random,
                                List<Tree> binarizedTrainTrees,
                                List<List<Transition>> transitionLists,
                                Treebank devTreebank,
                                int nThreads)

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.

Parameters:

serializedPath - Where serialized models go. If the appropriate options are set, the method can use this to save intermediate models.

tagger - The tagger to use when evaluating devTreebank. TODO: it would make more sense for ShiftReduceParser to retag the trees first

random - A random number generator to use for any random numbers. Useful to make sure results can be reproduced.

binarizedTrainTrees - The treebank to train from.

transitionLists - binarizedTrainTrees converted into lists of transitions that will reproduce the same tree.

devTreebank - a set of trees which can be used for dev testing (assuming the user provided a dev treebank)

nThreads - how many threads the model can use for training