Class XGBoostSuggestionsOrderer

java.lang.Object

org.languagetool.rules.spelling.suggestions.SuggestionsOrdererFeatureExtractor

org.languagetool.rules.spelling.suggestions.XGBoostSuggestionsOrderer

All Implemented Interfaces:

SuggestionsOrderer, SuggestionsRanker

public class XGBoostSuggestionsOrderer
extends SuggestionsOrdererFeatureExtractor
implements SuggestionsRanker

Nested Class Summary

Nested classes/interfaces inherited from class SuggestionsOrdererFeatureExtractor

SuggestionsOrdererFeatureExtractor.Feature

Field Summary

Fields

Modifier and Type	Field	Description
private static final Map<String,Float>	autoCorrectThreshold
private static final Map<String,Integer>	candidateFeatureCount
private static final org.slf4j.Logger	logger
private static final Map<String,Integer>	matchFeatureCount
private boolean	modelAvailableForLanguage
private static final Map<String, List<Integer>>	modelClasses
private static final org.apache.commons.pool2.KeyedObjectPool<Language, ml.dmlc.xgboost4j.java.Booster>	modelPool
private static boolean	xgboostNotSupported

Fields inherited from class SuggestionsOrdererFeatureExtractor

language, languageModel, mistakeProb, score, topN

Constructor Summary

Constructors

Constructor	Description
XGBoostSuggestionsOrderer(Language lang, LanguageModel languageModel)

Method Summary

Modifier and Type	Method	Description
private static @NotNull String	getModelPath(Language language)
protected void	initParameters()
boolean	isMlAvailable()
List<SuggestedReplacement>	orderSuggestions(List<String> suggestions, String word, AnalyzedSentence sentence, int startPos)
static void	setAutoCorrectThresholdForLanguage(Language lang, float value)	For testing purposes only
boolean	shouldAutoCorrect(List<SuggestedReplacement> rankedSuggestions)	Model output should have been calibrated using a precision-recall curve evaluation, so that a threshold for confidence values with sufficiently high precision for auto correction is known

Methods inherited from class SuggestionsOrdererFeatureExtractor

computeFeatures

Methods inherited from class Object

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

Methods inherited from interface SuggestionsOrderer

orderSuggestionsUsingModel

Field Details

logger

private static final org.slf4j.Logger logger

modelPool

private static final org.apache.commons.pool2.KeyedObjectPool<Language, ml.dmlc.xgboost4j.java.Booster> modelPool

autoCorrectThreshold

private static final Map<String,Float> autoCorrectThreshold

modelClasses

private static final Map<String, List<Integer>> modelClasses

candidateFeatureCount

private static final Map<String,Integer> candidateFeatureCount

matchFeatureCount

private static final Map<String,Integer> matchFeatureCount

modelAvailableForLanguage

private boolean modelAvailableForLanguage

xgboostNotSupported

private static boolean xgboostNotSupported

Constructor Details

XGBoostSuggestionsOrderer

public XGBoostSuggestionsOrderer(Language lang,
 LanguageModel languageModel)

Method Details

getModelPath

@NotNull
private static @NotNull String getModelPath(Language language)

setAutoCorrectThresholdForLanguage

public static void setAutoCorrectThresholdForLanguage(Language lang,
 float value)

For testing purposes only

initParameters

protected void initParameters()

Overrides:

initParameters in class SuggestionsOrdererFeatureExtractor

isMlAvailable

public boolean isMlAvailable()

Specified by:

isMlAvailable in interface SuggestionsOrderer

Overrides:

isMlAvailable in class SuggestionsOrdererFeatureExtractor

orderSuggestions

public List<SuggestedReplacement> orderSuggestions(List<String> suggestions,
 String word,
 AnalyzedSentence sentence,
 int startPos)

Specified by:

orderSuggestions in interface SuggestionsOrderer

Overrides:

orderSuggestions in class SuggestionsOrdererFeatureExtractor

shouldAutoCorrect

public boolean shouldAutoCorrect(List<SuggestedReplacement> rankedSuggestions)

Description copied from interface: SuggestionsRanker

Model output should have been calibrated using a precision-recall curve evaluation, so that a threshold for confidence values with sufficiently high precision for auto correction is known

Specified by:

shouldAutoCorrect in interface SuggestionsRanker

Parameters:

rankedSuggestions - suggestions returned by orderSuggestions

Returns:

if confidence is high enough for auto correction