private Features GetFeatures(List <FeatureTemplate> list, Token token, IndexedSentence sentence, bool?local = null)
{
Features features = new Features(list.Count);
foreach (FeatureTemplate template in list)
{
if (local == null || local == template.IsLocal)
{
features.Add(template.Name, template.GetValue(token, sentence));
}
}
return(features);
}
public void Tag(IEnumerable <Sentence> sentences)
{
if (perceptronMsd == null)
{
throw new InvalidOperationException("Model not loaded.");
}
Parallel.ForEach(sentences, (sentence, index) =>
{
IndexedSentence normalizedSentence =
new IndexedSentence((Reverse ? (sentence as IEnumerable <Token>).Reverse() : sentence).Select(t => Normalize(t)));
foreach (Token token in normalizedSentence)
{
Tag[] possibleTags = token.PossibleTags;
if (possibleTags == null || possibleTags.Length == 0)
{
possibleTags = perceptronMsd.Tags.ToArray();
}
if (possibleTags.Length == 1)
{
token.PredictedTag = possibleTags[0];
}
else
{
double?maxScoreMsd = null;
Tag bestMsd = null;
Features localFeaturesTag = GetFeatures(featureTemplatesTag, token, normalizedSentence, true);
foreach (Tag tag in possibleTags)
{
Tag tagMsd = new Tag(tag.Msd);
token.PredictedTag = tagMsd;
Features featuresTag = GetFeatures(featureTemplatesTag, token, normalizedSentence, false);
featuresTag.AddRange(localFeaturesTag);
double scoreMsd = perceptronMsd.Score(featuresTag, tagMsd);
if (maxScoreMsd == null || scoreMsd > maxScoreMsd.Value)
{
maxScoreMsd = scoreMsd;
bestMsd = tagMsd;
}
}
double?maxScoreLemma = null;
string bestLemma = null;
token.PredictedTag = bestMsd;
Features localFeaturesLemma = GetFeatures(featureTemplatesLemma, token, normalizedSentence, true);
foreach (Tag tag in possibleTags)
{
if (tag.Msd != bestMsd.Msd || tag.Lemma == null)
{
continue;
}
token.PredictedTag = new Tag(token.CorrectTag.Msd, tag.Lemma);
Features featuresLemma = GetFeatures(featureTemplatesLemma, token, normalizedSentence, false);
featuresLemma.AddRange(localFeaturesLemma);
double scoreLemma = perceptronLemma.Score(featuresLemma, tag.Lemma);
if (maxScoreLemma == null || scoreLemma > maxScoreLemma.Value)
{
maxScoreLemma = scoreLemma;
bestLemma = tag.Lemma;
}
}
token.PredictedTag = new Tag(bestMsd.Msd, bestLemma);
}
int pos = normalizedSentence[token];
if (Reverse)
{
pos = normalizedSentence.Count - pos - 1;
}
Token original = sentence[pos];
original.PredictedTag = token.PredictedTag;
Unnormalize(original);
}
});
}
public abstract string GetValue(Token token, IndexedSentence sentence);
