/// <summary>
/// Generates chunk tags for the given sequence returning the result in an array.
/// </summary>
/// <param name="tokens">an array of the tokens or words of the sequence.</param>
/// <param name="tags">an array of the pos tags of the sequence.</param>
/// <returns>An array of chunk tags for each token in the sequence or a <c>null</c> value if none.</returns>
/// <exception cref="System.ArgumentNullException">
/// The <paramref name="tokens"/> is null.
/// or
/// The <paramref name="tags"/> is null.
/// </exception>
/// <exception cref="System.ArgumentOutOfRangeException">The token array is empty.</exception>
public string[] Chunk(string[] tokens, string[] tags)
{
if (tokens == null)
{
throw new ArgumentNullException("tokens");
}
if (tokens.Length == 0)
{
throw new ArgumentOutOfRangeException("tokens", "The token array is empty.");
}
if (tags == null)
{
throw new ArgumentNullException("tags");
}
bestSequence = model.BestSequence(tokens, new object[] { tags }, contextGenerator, sequenceValidator);
return(bestSequence == null ? null : bestSequence.Outcomes.ToArray());
}
/// <summary>
/// Returns the lemma of the specified word with the specified part-of-speech.
/// </summary>
/// <param name="tokens">An array of the tokens.</param>
/// <param name="tags">An array of the POS tags.</param>
/// <returns>An array of lemma classes for each token in the sequence.</returns>
/// <exception cref="ArgumentNullException"><paramref name="tokens" /> or <paramref name="tags" /></exception>
/// <exception cref="ArgumentException">The arguments must have the same length.</exception>
public string[] Lemmatize(string[] tokens, string[] tags)
{
if (tokens == null)
{
throw new ArgumentNullException(nameof(tokens));
}
if (tags == null)
{
throw new ArgumentNullException(nameof(tags));
}
if (tokens.Length != tags.Length)
{
throw new ArgumentException("The arguments must have the same length.");
}
bestSequence = model.BestSequence(tokens, new object[] { tags }, contextGenerator, sequenceValidator);
return(bestSequence.Outcomes.ToArray());
}
/// <summary>
/// Generates name tags for the given sequence, typically a sentence, returning token spans for any identified names.
/// </summary>
/// <param name="tokens">An array of the tokens or words of the sequence, typically a sentence.</param>
/// <param name="additionalContext">Features which are based on context outside of the sentence but which should also be used.</param>
/// <returns>An array of spans for each of the names identified.</returns>
public Span[] Find(string[] tokens, string[][] additionalContext)
{
additionalContextFeatureGenerator.SetCurrentContext(additionalContext);
bestSequence = model.BestSequence(tokens,
Array.ConvertAll(additionalContext, input => (object)input),
contextGenerator,
sequenceValidator);
var outcomes = bestSequence.Outcomes.ToArray();
contextGenerator.UpdateAdaptiveData(tokens, outcomes);
var spans = sequenceCodec.Decode(outcomes);
var probs = Probs(spans);
for (var i = 0; i < probs.Length; i++)
{
spans[i].Probability = probs[i];
}
return(spans);
}
/// <summary>
/// Assigns the sentence of tokens pos tags.
/// </summary>
/// <param name="sentence">The sentence of tokens to be tagged.</param>
/// <param name="additionalContext">Any addition context specific to a class implementing this interface.</param>
/// <returns>an array of pos tags for each token provided in sentence.</returns>
public string[] Tag(string[] sentence, object[] additionalContext) {
bestSequence = model.BestSequence(sentence, additionalContext, ContextGenerator, SequenceValidator);
return bestSequence == null ? new string[0] : bestSequence.Outcomes.ToArray();
}
/// <summary>
/// Finds the n most probable sequences.
/// </summary>
/// <param name="numSequences">The number sequences.</param>
/// <param name="sequence">The sequence.</param>
/// <param name="additionalContext">The additional context.</param>
/// <param name="minSequenceScore">The minimum sequence score.</param>
/// <param name="beamSearch">The beam search.</param>
/// <param name="validator">The validator.</param>
public Sequence[] BestSequences(int numSequences, T[] sequence, object[] additionalContext,
double minSequenceScore,
IBeamSearchContextGenerator <T> beamSearch, ISequenceValidator <T> validator)
{
IHeap <Sequence> prev = new ListHeap <Sequence>(size);
IHeap <Sequence> next = new ListHeap <Sequence>(size);
prev.Add(new Sequence());
if (additionalContext == null)
{
additionalContext = new object[] {}; // EMPTY_ADDITIONAL_CONTEXT
}
for (var i = 0; i < sequence.Length; i++)
{
var sz = Math.Min(size, prev.Size());
for (var sc = 0; prev.Size() > 0 && sc < sz; sc++)
{
var top = prev.Extract();
var tmpOutcomes = top.Outcomes;
var outcomes = tmpOutcomes.ToArray();
var contexts = beamSearch.GetContext(i, sequence, outcomes, additionalContext);
double[] scores;
if (contextsCache != null)
{
scores = (double[])contextsCache.Get(contexts);
if (scores == null)
{
scores = model.Eval(contexts, probs);
contextsCache.Put(contexts, scores);
}
}
else
{
scores = model.Eval(contexts, probs);
}
var tempScores = new double[scores.Length];
for (var c = 0; c < scores.Length; c++)
{
tempScores[c] = scores[c];
}
Array.Sort(tempScores);
var min = tempScores[Math.Max(0, scores.Length - size)];
for (var p = 0; p < scores.Length; p++)
{
if (scores[p] < min)
{
continue; //only advance first "size" outcomes
}
var outcome = model.GetOutcome(p);
if (validator.ValidSequence(i, sequence, outcomes, outcome))
{
var ns = new Sequence(top, outcome, scores[p]);
if (ns.Score > minSequenceScore)
{
next.Add(ns);
}
}
}
if (next.Size() == 0)
{
//if no advanced sequences, advance all valid
for (var p = 0; p < scores.Length; p++)
{
var outcome = model.GetOutcome(p);
if (validator.ValidSequence(i, sequence, outcomes, outcome))
{
var ns = new Sequence(top, outcome, scores[p]);
if (ns.Score > minSequenceScore)
{
next.Add(ns);
}
}
}
}
}
// make prev = next; and re-init next (we reuse existing prev set once we clear it)
prev.Clear();
var tmp = prev;
prev = next;
next = tmp;
}
var numSeq = Math.Min(numSequences, prev.Size());
var topSequences = new Sequence[numSeq];
for (var seqIndex = 0; seqIndex < numSeq; seqIndex++)
{
topSequences[seqIndex] = prev.Extract();
}
return(topSequences);
}
/// <summary>
/// Generates chunk tags for the given sequence returning the result in an array.
/// </summary>
/// <param name="tokens">an array of the tokens or words of the sequence.</param>
/// <param name="tags">an array of the pos tags of the sequence.</param>
/// <returns>an array of chunk tags for each token in the sequence.</returns>
public string[] Chunk(string[] tokens, string[] tags)
{
bestSequence = model.BestSequence(tokens, new object[] { tags }, contextGenerator, sequenceValidator);
return(bestSequence.Outcomes.ToArray());
}
/// <summary>
/// Generates name tags for the given sequence, typically a sentence, returning token spans for any identified names.
/// </summary>
/// <param name="tokens">An array of the tokens or words of the sequence, typically a sentence.</param>
/// <param name="additionalContext">Features which are based on context outside of the sentence but which should also be used.</param>
/// <returns>An array of spans for each of the names identified.</returns>
public Span[] Find(string[] tokens, string[][] additionalContext) {
additionalContextFeatureGenerator.SetCurrentContext(additionalContext);
bestSequence = model.BestSequence(tokens,
Array.ConvertAll(additionalContext, input => (object) input),
contextGenerator,
sequenceValidator);
var outcomes = bestSequence.Outcomes.ToArray();
contextGenerator.UpdateAdaptiveData(tokens, outcomes);
var spans = sequenceCodec.Decode(outcomes);
var probs = Probs(spans);
for (var i = 0; i < probs.Length; i++) {
spans[i].Probability = probs[i];
}
return spans;
}
/// <summary>
/// Assigns the sentence of tokens pos tags.
/// </summary>
/// <param name="sentence">The sentence of tokens to be tagged.</param>
/// <param name="additionalContext">Any addition context specific to a class implementing this interface.</param>
/// <returns>an array of pos tags for each token provided in sentence.</returns>
public string[] Tag(string[] sentence, object[] additionalContext)
{
bestSequence = model.BestSequence(sentence, additionalContext, ContextGenerator, SequenceValidator);
return(bestSequence.Outcomes.ToArray());
}
/// <summary>
/// Generates chunk tags for the given sequence returning the result in an array.
/// </summary>
/// <param name="tokens">an array of the tokens or words of the sequence.</param>
/// <param name="tags">an array of the pos tags of the sequence.</param>
/// <returns>An array of chunk tags for each token in the sequence or a <c>null</c> value if none.</returns>
/// <exception cref="System.ArgumentNullException">
/// The <paramref name="tokens"/> is null.
/// or
/// The <paramref name="tags"/> is null.
/// </exception>
/// <exception cref="System.ArgumentOutOfRangeException">The token array is empty.</exception>
public string[] Chunk(string[] tokens, string[] tags) {
if (tokens == null)
throw new ArgumentNullException("tokens");
if (tokens.Length == 0)
throw new ArgumentOutOfRangeException("tokens", "The token array is empty.");
if (tags == null)
throw new ArgumentNullException("tags");
bestSequence = model.BestSequence(tokens, new object[] {tags}, contextGenerator, sequenceValidator);
return bestSequence == null ? null : bestSequence.Outcomes.ToArray();
}
