/// <summary>
/// Returns the next object. Calling this method repeatedly until it returns,
/// null will return each object from the underlying source exactly once.
/// </summary>
/// <returns>
/// The next object or null to signal that the stream is exhausted.
/// </returns>
public NameSample Read()
{
AdSentence paragraph;
if (monitor != null)
{
monitor.Token.ThrowIfCancellationRequested();
}
while ((paragraph = adSentenceStream.Read()) != null)
{
var clearData = false;
var currentTextID = GetTextId(paragraph);
if (currentTextID != textId)
{
clearData = true;
textId = currentTextID;
}
var root = paragraph.Root;
var sentence = new List <string>();
var names = new List <Span>();
Process(root, sentence, names);
return(new NameSample(sentence.ToArray(), names.ToArray(), clearData));
}
return(null);
}
/// <summary>
/// Returns the next object. Calling this method repeatedly until it returns,
/// null will return each object from the underlying source exactly once.
/// </summary>
/// <returns>
/// The next object or null to signal that the stream is exhausted.
/// </returns>
public Sequence Read()
{
var sample = psi.Read();
if (sample != null)
{
var events = new Event[sample.Sentence.Length];
for (int i = 0; i < sample.Sentence.Length; i++)
{
// it is safe to pass the tags as previous tags because
// the context generator does not look for non predicted tags
var tags = seqCodec.Encode(sample.Names, sample.Sentence.Length);
var context = pcg.GetContext(
i,
sample.Sentence,
useOutcomes ? tags : null,
null);
events[i] = new Event(tags[i], context);
}
return(new Sequence(events, sample));
}
return(null);
}
private List <ComparableEvent> Index(
IObjectStream <Event> indexEventStream,
Dictionary <string, int> predicateIndex)
{
var map = new Dictionary <string, int>();
var indexedContext = new List <int>();
var eventsToCompare = new List <ComparableEvent>();
int outcomeCount = 0;
Event ev;
while ((ev = indexEventStream.Read()) != null)
{
int ocId;
if (Monitor != null && Monitor.Token.CanBeCanceled)
{
Monitor.Token.ThrowIfCancellationRequested();
}
if (map.ContainsKey(ev.Outcome))
{
ocId = map[ev.Outcome];
}
else
{
ocId = outcomeCount++;
map[ev.Outcome] = ocId;
}
// ReSharper disable once LoopCanBeConvertedToQuery
foreach (var pred in ev.Context)
{
if (predicateIndex.ContainsKey(pred))
{
indexedContext.Add(predicateIndex[pred]);
}
}
// drop events with no active features
if (indexedContext.Count > 0)
{
var cons = new int[indexedContext.Count];
for (int ci = 0; ci < cons.Length; ci++)
{
cons[ci] = indexedContext[ci];
}
eventsToCompare.Add(new ComparableEvent(ocId, cons));
}
else
{
Monitor?.OnWarning($"Dropped event {ev.Outcome}:{ev.Context.ToDisplay()}");
}
indexedContext.Clear();
}
outcomeLabels = ToIndexedStringArray(map);
predLabels = ToIndexedStringArray(predicateIndex);
return(eventsToCompare);
}
/// <summary>
/// Returns the next object. Calling this method repeatedly until it returns,
/// null will return each object from the underlying source exactly once.
/// </summary>
/// <returns>
/// The next object or null to signal that the stream is exhausted.
/// </returns>
public ChunkSample Read()
{
AdSentence paragraph;
while ((paragraph = adSentenceStream.Read()) != null)
{
if (End > -1 && Index >= End) // leave
{
return(null);
}
if (Start > -1 && Index < Start)
{
Index++;
}
else
{
var root = paragraph.Root;
var sentence = new List <string>();
var tags = new List <string>();
var target = new List <string>();
ProcessRoot(root, sentence, tags, target);
if (sentence.Count <= 0)
{
continue;
}
Index++;
return(new ChunkSample(sentence.ToArray(), tags.ToArray(), target.ToArray()));
}
}
return(null);
}
/// <summary>
/// Returns the next object. Calling this method repeatedly until it returns,
/// null will return each object from the underlying source exactly once.
/// </summary>
/// <returns>
/// The next object or null to signal that the stream is exhausted.
/// </returns>
public Event Read()
{
var eventString = objectStream.Read();
return(eventString != null
? CreateEvent(eventString, Monitor)
: null);
}
public static void PopulatePOSDictionary(IObjectStream <POSSample> samples, IMutableTagDictionary dictionary, bool caseSensitive, int cutoff)
{
var newEntries = new Dictionary <string, Dictionary <string, int> >();
POSSample sample;
while ((sample = samples.Read()) != null)
{
for (int i = 0; i < sample.Sentence.Length; i++)
{
if (!StringPattern.Recognize(sample.Sentence[i]).ContainsDigit)
{
string word = caseSensitive ? sample.Sentence[i] : sample.Sentence[i].ToLowerInvariant();
if (!newEntries.ContainsKey(word))
{
newEntries.Add(word, new Dictionary <string, int>());
}
var dicTags = dictionary.GetTags(word);
if (dicTags != null)
{
foreach (var tag in dicTags)
{
if (!newEntries[word].ContainsKey(tag))
{
newEntries[word].Add(tag, cutoff);
}
}
}
if (!newEntries[word].ContainsKey(sample.Tags[i]))
{
newEntries[word].Add(sample.Tags[i], 1);
}
else
{
newEntries[word][sample.Tags[i]]++;
}
}
}
}
foreach (var wordEntry in newEntries)
{
var tagsForWord = new List <string>();
foreach (var entry in wordEntry.Value)
{
if (entry.Value >= cutoff)
{
tagsForWord.Add(entry.Key);
}
}
if (tagsForWord.Count > 0)
{
dictionary.Put(wordEntry.Key, tagsForWord.ToArray());
}
}
}
public T Read()
{
if (isPoisoned)
{
throw new InvalidOperationException();
}
// skip training samples
while (index % numberOfPartitions != testIndex)
{
sampleStream.Read();
index++;
}
index++;
return(sampleStream.Read());
}
/// <summary>
/// Reads all sample objects from the stream and evaluates each sample object with <see cref="M:ProcessSample"/> method.
/// </summary>
/// <param name="samples">The samples.</param>
public void Evaluate(IObjectStream <T> samples)
{
T sample;
while ((sample = samples.Read()) != null)
{
EvaluateSample(sample);
}
}
/// <summary>
/// Returns the next object. Calling this method repeatedly until it returns ,
/// null will return each object from the underlying source exactly once.
/// </summary>
/// <returns>
/// The next object or null to signal that the stream is exhausted.
/// </returns>
public T Read()
{
if (testSampleStream != null || isPoisoned)
{
throw new InvalidOperationException();
}
// If the test element is reached skip over it to not include it in
// the training data
if (index % numberOfPartitions == testIndex)
{
sampleStream.Read();
index++;
}
index++;
return(sampleStream.Read());
}
/// <summary>
/// Returns the next object. Calling this method repeatedly until it returns, null will return each object from the underlying source exactly once.
/// </summary>
/// <returns>
/// The next object or null to signal that the stream is exhausted.
/// </returns>
public virtual T Read()
{
var value = stream.Read();
if (value == null)
{
IsFinished = true;
}
return(value);
}
/// <summary>
/// Returns the next object. Calling this method repeatedly until it returns,
/// null will return each object from the underlying source exactly once.
/// </summary>
/// <returns>
/// The next object or null to signal that the stream is exhausted.
/// </returns>
public TokenSample Read()
{
var inputString = input.Read();
if (inputString != null)
{
var tokens = tokenizer.TokenizePos(inputString);
return(new TokenSample(inputString, tokens));
}
return(null);
}
/// <summary>
/// Returns the next <see cref="T:SentenceSample"/>. Calling this method repeatedly until it returns,
/// null will return each object from the underlying source exactly once.
/// </summary>
/// <returns>
/// The next object or null to signal that the stream is exhausted.
/// </returns>
public SentenceSample Read()
{
if (sentence == null)
{
sentence = adSentenceStream.Read();
UpdateMeta();
if (sentence == null)
{
return(null);
}
}
var document = new StringBuilder();
var sentences = new List <Span>();
do
{
do
{
if (!isTitle || (isTitle && isIncludeTitles))
{
if (HasPunctuation(sentence.Text))
{
var start = document.Length;
document.Append(sentence.Text);
sentences.Add(new Span(start, document.Length));
document.Append(' ');
}
}
sentence = adSentenceStream.Read();
UpdateMeta();
} while (isSamePara);
} while (isSameText);
return(new SentenceSample(
document.Length > 0 ? document.ToString(0, document.Length - 1) : document.ToString(),
sentences.ToArray()
));
}
/// <summary>
/// Returns the next object. Calling this method repeatedly until it returns,
/// null will return each object from the underlying source exactly once.
/// </summary>
/// <returns>
/// The next object or null to signal that the stream is exhausted.
/// </returns>
public POSSample Read()
{
callsCount++;
AdSentence paragraph;
while ((paragraph = adSentenceStream.Read()) != null)
{
var root = paragraph.Root;
var sentence = new List <string>();
var tags = new List <string>();
var contractions = new List <string>();
var prop = new List <string>();
Process(root, sentence, tags, contractions, prop);
if (sentence.Count != contractions.Count || sentence.Count != prop.Count)
{
throw new InvalidOperationException("The processed information must have the same length.");
}
if (additionalContext)
{
//String[][] ac = new String[2][sentence.size()];
var ac = new string[2][];
ac[0] = new string[sentence.Count];
ac[1] = new string[sentence.Count];
// line 0: contractions
// line 1: props
for (var i = 0; i < sentence.Count; i++)
{
if (contractions[i] != null)
{
ac[0][i] = contractions[i];
}
if (prop[i] != null)
{
ac[1][i] = prop[i];
}
}
return(new POSSample(sentence.ToArray(), tags.ToArray(), ac));
}
return(new POSSample(sentence.ToArray(), tags.ToArray()));
}
return(null);
}
/// <summary>
/// Builds the NGram dictionary with the given samples.
/// </summary>
/// <param name="samples">The samples.</param>
/// <param name="cutoff">The cutoff.</param>
/// <returns>The NGram dictionary.</returns>
public static Dic BuildNGramDictionary(IObjectStream <POSSample> samples, int cutoff)
{
var model = new NGramModel();
POSSample sample;
while ((sample = samples.Read()) != null)
{
if (sample.Sentence.Length > 0)
{
model.Add(new StringList(sample.Sentence), 1, 1);
}
}
model.CutOff(cutoff, int.MaxValue);
return(model.ToDictionary());
}
/// <summary>
/// Returns the next <see cref="Sequence"/>. Calling this method repeatedly until it returns,
/// null will return each object from the underlying source exactly once.
/// </summary>
/// <returns>
/// The next <see cref="Sequence"/> or null to signal that the stream is exhausted.
/// </returns>
public Sequence Read()
{
var sample = objectStream.Read();
if (sample != null)
{
var events = new Event[sample.Sentence.Length];
for (var i = 0; i < events.Length; i++)
{
events[i] = new Event(sample.Tags[i],
contextGenerator.GetContext(i, sample.Sentence, sample.Tags, null));
}
return(new Sequence(events, sample));
}
return(null);
}
/// <summary>
/// Returns the next <see cref="Sequence"/>. Calling this method repeatedly until it returns,
/// null will return each object from the underlying source exactly once.
/// </summary>
/// <returns>
/// The next object or null to signal that the stream is exhausted.
/// </returns>
public Sequence Read()
{
var sample = samples.Read();
if (sample != null)
{
var events = new Event[sample.Sentence.Count];
for (int i = 0, count = sample.Sentence.Count; i < count; i++)
{
events[i] = new Event(
sample.Tags[i],
// it is safe to pass the tags as previous tags because
// the context generator does not look for non predicted tags
contextGenerator.GetContext(i, sample.Sentence.ToArray(), sample.Tags.ToArray(), null));
}
return(new Sequence(events, sample));
}
return(null);
}
/// <summary>
/// Returns the next <see cref="T:Event"/>. Calling this method repeatedly until it returns,
/// null will return each event from the underlying source exactly once.
/// </summary>
/// <returns>
/// The next object or null to signal that the stream is exhausted.
/// </returns>
public Event Read()
{
if (events.MoveNext())
{
return(events.Current);
}
T sample;
while ((sample = samples.Read()) != null)
{
events = CreateEvents(sample);
if (events != null && events.MoveNext())
{
return(events.Current);
}
}
return(null);
}
/// <summary>
/// Returns the next object. Calling this method repeatedly until it returns,
/// null will return each object from the underlying source exactly once.
/// </summary>
/// <returns>
/// The next object or null to signal that the stream is exhausted.
/// </returns>
public NameSample Read()
{
AdSentence paragraph;
while ((paragraph = adSentenceStream.Read()) != null)
{
if (monitor != null)
{
monitor.Token.ThrowIfCancellationRequested();
}
var root = paragraph.Root;
var sentence = new List <string>();
var names = new List <Span>();
Process(root, sentence, names);
return(new NameSample(sentence.ToArray(), names.ToArray(), true));
}
return(null);
}
public Sequence Read()
{
var sample = samples.Read();
if (sample == null)
{
return(null);
}
var events = new Event[sample.Length];
for (var i = 0; i < sample.Length; i++)
{
// it is safe to pass the tags as previous tags because
// the context generator does not look for non predicted tags
var context = contextGenerator.GetContext(i, sample.Tokens, sample.Tags, sample.Lemmas);
events[i] = new Event(sample.Tags[i], context);
}
return(new Sequence(events, sample));
}
/// <summary>
/// Creates a n-gram dictionary from the specified data stream using the specified head rule and specified cut-off.
/// </summary>
/// <param name="data">The data stream of parses.</param>
/// <param name="rules">The head rules for the parses.</param>
/// <param name="parameters">Can contain a cutoff, the minimum number of entries required for the n-gram to be saved as part of the dictionary.</param>
/// <returns>A dictionary object.</returns>
public static Dic BuildDictionary(IObjectStream<Parse> data, AbstractHeadRules rules, TrainingParameters parameters) {
var cutoff = parameters.Get("dict", Parameters.Cutoff, 5);
var dict = new NGramModel();
Parse p;
while ((p = data.Read()) != null) {
p.UpdateHeads(rules);
var pWords = p.GetTagNodes();
var words = new string[pWords.Length];
//add all uni-grams
for (var wi = 0; wi < words.Length; wi++) {
words[wi] = pWords[wi].CoveredText;
}
dict.Add(new StringList(words), 1, 1);
//add tri-grams and bi-grams for initial sequence
var chunks = CollapsePunctuation(AbstractParserEventStream.GetInitialChunks(p), rules.PunctuationTags);
var cWords = new string[chunks.Length];
for (var wi = 0; wi < cWords.Length; wi++) {
cWords[wi] = chunks[wi].Head.CoveredText;
}
dict.Add(new StringList(cWords), 2, 3);
//emulate reductions to produce additional n-grams
var ci = 0;
while (ci < chunks.Length) {
/*
if (chunks[ci].Parent == null) {
chunks[ci].Show();
} */
if (LastChild(chunks[ci], chunks[ci].Parent, rules.PunctuationTags)) {
//perform reduce
var reduceStart = ci;
while (reduceStart >= 0 && Equals(chunks[reduceStart].Parent, chunks[ci].Parent)) {
reduceStart--;
}
reduceStart++;
chunks = ParserEventStream.ReduceChunks(chunks, ref ci, chunks[ci].Parent);
ci = reduceStart;
if (chunks.Length != 0) {
var window = new string[5];
var wi = 0;
if (ci - 2 >= 0) window[wi++] = chunks[ci - 2].Head.CoveredText;
if (ci - 1 >= 0) window[wi++] = chunks[ci - 1].Head.CoveredText;
window[wi++] = chunks[ci].Head.CoveredText;
if (ci + 1 < chunks.Length) window[wi++] = chunks[ci + 1].Head.CoveredText;
if (ci + 2 < chunks.Length) window[wi++] = chunks[ci + 2].Head.CoveredText;
if (wi < 5) {
var subWindow = new string[wi];
for (var swi = 0; swi < wi; swi++) {
subWindow[swi] = window[swi];
}
window = subWindow;
}
if (window.Length >= 3) {
dict.Add(new StringList(window), 2, 3);
} else if (window.Length == 2) {
dict.Add(new StringList(window), 2, 2);
}
}
ci = reduceStart - 1; //ci will be incremented at end of loop
}
ci++;
}
}
dict.CutOff(cutoff, int.MaxValue);
return dict.ToDictionary(true);
}
private List<ComparableEvent> Index(
IObjectStream<Event> indexEventStream,
Dictionary<string, int> predicateIndex) {
var map = new Dictionary<string, int>();
var indexedContext = new List<int>();
var eventsToCompare = new List<ComparableEvent>();
int outcomeCount = 0;
Event ev;
while ((ev = indexEventStream.Read()) != null) {
int ocID;
if (Monitor != null && Monitor.Token.CanBeCanceled)
Monitor.Token.ThrowIfCancellationRequested();
if (map.ContainsKey(ev.Outcome)) {
ocID = map[ev.Outcome];
} else {
ocID = outcomeCount++;
map[ev.Outcome] = ocID;
}
// ReSharper disable once LoopCanBeConvertedToQuery
foreach (var pred in ev.Context) {
if (predicateIndex.ContainsKey(pred)) {
indexedContext.Add(predicateIndex[pred]);
}
}
// drop events with no active features
if (indexedContext.Count > 0) {
var cons = new int[indexedContext.Count];
for (int ci = 0; ci < cons.Length; ci++) {
cons[ci] = indexedContext[ci];
}
eventsToCompare.Add(new ComparableEvent(ocID, cons));
} else {
if (Monitor != null)
Monitor.OnWarning(string.Format("Dropped event {0}:{1}", ev.Outcome, ev.Context.ToDisplay()));
}
indexedContext.Clear();
}
outcomeLabels = ToIndexedStringArray(map);
predLabels = ToIndexedStringArray(predicateIndex);
return eventsToCompare;
}
public static void PopulatePOSDictionary(IObjectStream<POSSample> samples, IMutableTagDictionary dictionary, bool caseSensitive, int cutoff) {
var newEntries = new Dictionary<string, Dictionary<string, int>>();
POSSample sample;
while ((sample = samples.Read()) != null) {
for (int i = 0; i < sample.Sentence.Length; i++) {
if (!StringPattern.Recognize(sample.Sentence[i]).ContainsDigit) {
string word = caseSensitive ? sample.Sentence[i] : sample.Sentence[i].ToLowerInvariant();
if (!newEntries.ContainsKey(word)) {
newEntries.Add(word, new Dictionary<string, int>());
}
var dicTags = dictionary.GetTags(word);
if (dicTags != null) {
foreach (var tag in dicTags) {
if (!newEntries[word].ContainsKey(tag)) {
newEntries[word].Add(tag, cutoff);
}
}
}
if (!newEntries[word].ContainsKey(sample.Tags[i])) {
newEntries[word].Add(sample.Tags[i], 1);
} else {
newEntries[word][sample.Tags[i]]++;
}
}
}
}
foreach (var wordEntry in newEntries) {
var tagsForWord = (from entry in wordEntry.Value where entry.Value >= cutoff select entry.Key).ToList();
if (tagsForWord.Count > 0)
dictionary.Put(wordEntry.Key, tagsForWord.ToArray());
}
}
/// <summary>
/// Builds the NGram dictionary with the given samples.
/// </summary>
/// <param name="samples">The samples.</param>
/// <param name="cutoff">The cutoff.</param>
/// <returns>The NGram dictionary.</returns>
public static Dict BuildNGramDictionary(IObjectStream<POSSample> samples, int cutoff) {
var model = new NGramModel();
POSSample sample;
while ((sample = samples.Read()) != null) {
if (sample.Sentence.Length > 0) {
model.Add(new StringList(sample.Sentence), 1, 1);
}
}
model.CutOff(cutoff, int.MaxValue);
return model.ToDictionary();
}
/// <summary>
/// Creates a n-gram dictionary from the specified data stream using the specified head rule and specified cut-off.
/// </summary>
/// <param name="data">The data stream of parses.</param>
/// <param name="rules">The head rules for the parses.</param>
/// <param name="parameters">Can contain a cutoff, the minimum number of entries required for the n-gram to be saved as part of the dictionary.</param>
/// <returns>A dictionary object.</returns>
public static Dic BuildDictionary(IObjectStream <Parse> data, AbstractHeadRules rules, TrainingParameters parameters)
{
var cutoff = parameters.Get("dict", Parameters.Cutoff, 5);
var dict = new NGramModel();
Parse p;
while ((p = data.Read()) != null)
{
p.UpdateHeads(rules);
var pWords = p.GetTagNodes();
var words = new string[pWords.Length];
//add all uni-grams
for (var wi = 0; wi < words.Length; wi++)
{
words[wi] = pWords[wi].CoveredText;
}
dict.Add(new StringList(words), 1, 1);
//add tri-grams and bi-grams for initial sequence
var chunks = CollapsePunctuation(AbstractParserEventStream.GetInitialChunks(p), rules.PunctuationTags);
var cWords = new string[chunks.Length];
for (var wi = 0; wi < cWords.Length; wi++)
{
cWords[wi] = chunks[wi].Head.CoveredText;
}
dict.Add(new StringList(cWords), 2, 3);
//emulate reductions to produce additional n-grams
var ci = 0;
while (ci < chunks.Length)
{
/*
* if (chunks[ci].Parent == null) {
* chunks[ci].Show();
* } */
if (LastChild(chunks[ci], chunks[ci].Parent, rules.PunctuationTags))
{
//perform reduce
var reduceStart = ci;
while (reduceStart >= 0 && Equals(chunks[reduceStart].Parent, chunks[ci].Parent))
{
reduceStart--;
}
reduceStart++;
chunks = ParserEventStream.ReduceChunks(chunks, ref ci, chunks[ci].Parent);
ci = reduceStart;
if (chunks.Length != 0)
{
var window = new string[5];
var wi = 0;
if (ci - 2 >= 0)
{
window[wi++] = chunks[ci - 2].Head.CoveredText;
}
if (ci - 1 >= 0)
{
window[wi++] = chunks[ci - 1].Head.CoveredText;
}
window[wi++] = chunks[ci].Head.CoveredText;
if (ci + 1 < chunks.Length)
{
window[wi++] = chunks[ci + 1].Head.CoveredText;
}
if (ci + 2 < chunks.Length)
{
window[wi++] = chunks[ci + 2].Head.CoveredText;
}
if (wi < 5)
{
var subWindow = new string[wi];
for (var swi = 0; swi < wi; swi++)
{
subWindow[swi] = window[swi];
}
window = subWindow;
}
if (window.Length >= 3)
{
dict.Add(new StringList(window), 2, 3);
}
else if (window.Length == 2)
{
dict.Add(new StringList(window), 2, 2);
}
}
ci = reduceStart - 1; //ci will be incremented at end of loop
}
ci++;
}
}
dict.CutOff(cutoff, int.MaxValue);
return(dict.ToDictionary(true));
}
/// <summary>
/// Returns the next object. Calling this method repeatedly until it returns,
/// null will return each object from the underlying source exactly once.
/// </summary>
/// <returns>
/// The next object or null to signal that the stream is exhausted.
/// </returns>
public NameSample Read()
{
var sentence = new List <string>();
var tags = new List <string>();
var isClearAdaptiveData = false;
// Empty line indicates end of sentence
string line;
while ((line = lineStream.Read()) != null && !string.IsNullOrWhiteSpace(line))
{
if (line.StartsWith("###MEDLINE:"))
{
isClearAdaptiveData = true;
lineStream.Read();
continue;
}
if (line.Contains("ABSTRACT TRUNCATED"))
{
continue;
}
var fields = line.Split('\t');
if (fields.Length == 2)
{
sentence.Add(fields[0]);
tags.Add(fields[1]);
}
else
{
throw new InvalidFormatException("Expected two fields per line in training data, got " +
fields.Length + " for line '" + line + "'!");
}
}
if (sentence.Count > 0)
{
// convert name tags into spans
var names = new List <Span>();
var beginIndex = -1;
var endIndex = -1;
for (var i = 0; i < tags.Count; i++)
{
var tag = tags[i];
if (tag.EndsWith("DNA") && (types & GENERATE_DNA_ENTITIES) == 0)
{
tag = "O";
}
if (tag.EndsWith("protein") && (types & GENERATE_PROTEIN_ENTITIES) == 0)
{
tag = "O";
}
if (tag.EndsWith("cell_type") && (types & GENERATE_CELLTYPE_ENTITIES) == 0)
{
tag = "O";
}
if (tag.EndsWith("cell_line") && (types & GENERATE_CELLTYPE_ENTITIES) == 0)
{
tag = "O";
}
if (tag.EndsWith("RNA") && (types & GENERATE_RNA_ENTITIES) == 0)
{
tag = "O";
}
if (tag.StartsWith("B-"))
{
if (beginIndex != -1)
{
names.Add(new Span(beginIndex, endIndex, tags[beginIndex].Substring(2)));
//beginIndex = -1;
//endIndex = -1;
}
beginIndex = i;
endIndex = i + 1;
}
else if (tag.StartsWith("I-"))
{
endIndex++;
}
else if (tag.Equals("O"))
{
if (beginIndex != -1)
{
names.Add(new Span(beginIndex, endIndex, tags[beginIndex].Substring(2)));
beginIndex = -1;
endIndex = -1;
}
}
else
{
throw new IOException("Invalid tag: " + tag);
}
}
// if one span remains, create it here
if (beginIndex != -1)
{
names.Add(new Span(beginIndex, endIndex, tags[beginIndex].Substring(2)));
}
return(new NameSample(sentence.ToArray(), names.ToArray(), isClearAdaptiveData));
}
if (line != null)
{
// Just filter out empty events, if two lines in a row are empty
return(Read());
}
// source stream is not returning anymore lines
return(null);
}
/// <summary>
/// Returns the next object. Calling this method repeatedly until it returns,
/// null will return each object from the underlying source exactly once.
/// </summary>
/// <returns>
/// The next object or null to signal that the stream is exhausted.
/// </returns>
public NameSample Read()
{
var sentence = new List <string>();
var tags = new List <string>();
var ClearAdaptiveData = false;
// Empty line indicates end of sentence
string line;
while ((line = lineStream.Read()) != null && !string.IsNullOrWhiteSpace(line))
{
if (language == Language.Nl && line.StartsWith(DocStart))
{
ClearAdaptiveData = true;
continue;
}
var fields = line.Split(new[] { ' ' }, StringSplitOptions.RemoveEmptyEntries);
if (fields.Length == 3)
{
sentence.Add(fields[0]);
tags.Add(fields[2]);
}
else
{
throw new InvalidFormatException(
string.Format("Expected three fields per line in training data, got {0} for line '{1}'!",
fields.Length, line));
}
}
// Always clear adaptive data for spanish
if (language == Language.Es)
{
ClearAdaptiveData = true;
}
if (sentence.Count > 0)
{
// convert name tags into spans
var names = new List <Span>();
var beginIndex = -1;
var endIndex = -1;
for (var i = 0; i < tags.Count; i++)
{
var tag = tags[i];
if (tag.EndsWith("PER") && (types & Types.PersonEntities) == 0)
{
tag = "O";
}
if (tag.EndsWith("ORG") && (types & Types.OrganizationEntities) == 0)
{
tag = "O";
}
if (tag.EndsWith("LOC") && (types & Types.LocationEntities) == 0)
{
tag = "O";
}
if (tag.EndsWith("MISC") && (types & Types.MiscEntities) == 0)
{
tag = "O";
}
if (tag.StartsWith("B-"))
{
if (beginIndex != -1)
{
names.Add(Extract(beginIndex, endIndex, tags[beginIndex]));
//beginIndex = -1;
//endIndex = -1;
}
beginIndex = i;
endIndex = i + 1;
}
else if (tag.StartsWith("I-"))
{
endIndex++;
}
else if (tag.Equals("O"))
{
if (beginIndex != -1)
{
names.Add(Extract(beginIndex, endIndex, tags[beginIndex]));
beginIndex = -1;
endIndex = -1;
}
}
else
{
throw new InvalidFormatException("Invalid tag: " + tag);
}
}
// if one span remains, create it here
if (beginIndex != -1)
{
names.Add(Extract(beginIndex, endIndex, tags[beginIndex]));
}
return(new NameSample(sentence.ToArray(), names.ToArray(), ClearAdaptiveData));
}
return(line != null?Read() : null);
}
/// <summary>
/// Returns the next <see cref="PtbNode"/> object. Calling this method repeatedly until it returns,
/// null will return each object from the underlying source exactly once.
/// </summary>
/// <returns>
/// The next <see cref="PtbNode"/> object or <c>null</c> to signal that the stream is exhausted.
/// </returns>
public PtbNode Read()
{
retry:
if (monitor != null && monitor.Token.CanBeCanceled)
{
monitor.Token.ThrowIfCancellationRequested();
}
PtbNode root = null;
var pos = 0; // text position
var stack = new Stack <PtbNode>();
var invalid = false;
string line;
while ((line = lineStream.Read()) != null)
{
line = line.TrimStart(' ', '\t', '\u00A0'); // \u00A0 = NO-BREAK SPACE
if (invalid)
{
if (line.Trim().Length == 0) // end of sentence
{
goto retry;
}
for (var i = 0; i < line.Length; i++)
{
switch (line[i])
{
case '#':
if (i == 0)
{
goto next; // ignore comment
}
break;
case '(':
stack.Push(null);
continue;
case ')':
stack.Pop();
if (stack.Count == 0)
{
goto done;
}
continue;
default:
continue;
}
}
continue; // ignore invalid data
}
if (line.Length == 0)
{
if (root != null)
{
goto done;
}
continue;
}
for (var i = 0; i < line.Length; i++)
{
switch (line[i])
{
case '#':
if (i == 0)
{
goto next; // ignore comment, if the line starts with '#'
}
continue;
case '(':
var rest = line.Substring(i + 1);
var type = resolver.GetType(rest, useFunctionTags);
var token = resolver.GetToken(rest);
if (type == null)
{
if (monitor != null)
{
monitor.OnWarning("Penn treebank node without type: " + line);
}
stack.Push(null);
invalid = true;
goto next;
}
/* skip a few chars to improve performance (if possible)... */
int skip;
if (token != null && (skip = rest.IndexOf(')')) != -1)
{
i += skip;
}
var child = token != null
? new PtbNode {
Type = type, Token = token, Span = new Span(pos, pos + token.Length)
}
: new PtbNode {
Type = type
};
if (token != null)
{
pos += token.Length + 1;
}
if (root == null)
{
root = child;
}
if (stack.Count > 0)
{
var parent = stack.Peek();
// check if the parent node is a gap
if (parent != null)
{
parent.Children.Add(child);
}
else
{
// search for the parent node that is not a gap
var array = stack.ToArray();
foreach (var p in array)
{
if (p == null)
{
continue;
}
p.Children.Add(child);
break;
}
}
}
stack.Push(child);
continue;
case ')':
var pop = stack.Pop();
// adjust span
if (pop != null)
{
if (pop.HasChildren)
{
var s = GetStartPos(pop);
var e = GetEndPos(pop);
if (s.HasValue && e.HasValue)
{
pop.Span = new Span(s.Value, e.Value);
}
}
if (pop.Span == null)
{
pop.Span = null;
}
}
if (stack.Count == 0)
{
goto done;
}
continue;
}
}
next:
;
}
done:
// check if invalid.
if (invalid || stack.Count != 0)
{
if (monitor != null)
{
monitor.OnWarning("A invalid Penn Treebank sentence was skipped.");
}
goto retry;
}
// End of stream
if (root == null)
{
return(null);
}
var rs = GetStartPos(root);
var re = GetEndPos(root);
root.Span = new Span(rs.Value, re.Value);
// if the stack is not empty, the sentence is incomplete/invalid.
return(stack.Count == 0
? root
: null);
}
/// <summary>
/// Returns the next object. Calling this method repeatedly until it returns,
/// null will return each object from the underlying source exactly once.
/// </summary>
/// <returns>
/// The next object or null to signal that the stream is exhausted.
/// </returns>
public NameSample Read()
{
var sentence = new List <string>();
var tags = new List <string>();
var isClearAdaptiveData = false;
// Empty line indicates end of sentence
string line;
while ((line = lineStream.Read()) != null && !string.IsNullOrWhiteSpace(line))
{
if (line.StartsWith(DocStart))
{
isClearAdaptiveData = true;
line = lineStream.Read();
if (!string.IsNullOrEmpty(line))
{
throw new InvalidFormatException("Empty line after -DOCSTART- not empty: '" + line + "'!");
}
continue;
}
var fields = line.Split(new[] { ' ' }, StringSplitOptions.RemoveEmptyEntries);
if (language == Language.En && fields.Length == 4)
{
// For English: WORD POS-TAG SC-TAG NE-TAG
sentence.Add(fields[0]);
tags.Add(fields[3]);
}
else if (language == Language.De && fields.Length == 5)
{
// For German: WORD LEMA-TAG POS-TAG SC-TAG NE-TAG
sentence.Add(fields[0]);
tags.Add(fields[4]);
}
else
{
throw new InvalidFormatException(
$"Incorrect number of fields per line for language: '{line}'!");
}
}
if (sentence.Count > 0)
{
// convert name tags into spans
var names = new List <Span>();
var beginIndex = -1;
var endIndex = -1;
for (var i = 0; i < tags.Count; i++)
{
var tag = tags[i];
if (tag.EndsWith("PER") && (types & Types.PersonEntities) == 0)
{
tag = "O";
}
if (tag.EndsWith("ORG") && (types & Types.OrganizationEntities) == 0)
{
tag = "O";
}
if (tag.EndsWith("LOC") && (types & Types.LocationEntities) == 0)
{
tag = "O";
}
if (tag.EndsWith("MISC") && (types & Types.MiscEntities) == 0)
{
tag = "O";
}
if (tag == "O")
{
if (beginIndex == -1)
{
continue;
}
names.Add(Extract(beginIndex, endIndex, tags[beginIndex]));
beginIndex = -1;
endIndex = -1;
}
else if (tag.StartsWith("B-"))
{
// B- prefix means we have two same entities next to each other
if (beginIndex != -1)
{
names.Add(Extract(beginIndex, endIndex, tags[beginIndex]));
}
beginIndex = i;
endIndex = i + 1;
}
else if (tag.StartsWith("I-"))
{
// I- starts or continues a current name entity
if (beginIndex == -1)
{
beginIndex = i;
endIndex = i + 1;
}
else if (!tag.EndsWith(tags[beginIndex].Substring(1)))
{
// we have a new tag type following a tagged word series
// also may not have the same I- starting the previous!
names.Add(Extract(beginIndex, endIndex, tags[beginIndex]));
beginIndex = i;
endIndex = i + 1;
}
else
{
endIndex++;
}
}
else
{
throw new InvalidFormatException("Invalid tag: " + tag);
}
}
// if one span remains, create it here
if (beginIndex != -1)
{
names.Add(Extract(beginIndex, endIndex, tags[beginIndex]));
}
return(new NameSample(sentence.ToArray(), names.ToArray(), isClearAdaptiveData));
}
if (line != null)
{
// Just filter out empty events, if two lines in a row are empty
return(Read());
}
// source stream is not returning anymore lines
return(null);
}
