public virtual void TestRead()
{
string treeText = "(1 (2 This)) (3 (4 is) (5 a)) (6 (\\* small) (7 \\/test))";
StringReader reader = new StringReader(treeText);
PennTreeReader treeReader = new PennTreeReader(reader);
string[] expected = new string[] { "(1 (2 This))", "(3 (4 is) (5 a))", "(6 (* small) (7 /test))" };
for (int i = 0; i < expected.Length; ++i)
{
Tree tree = treeReader.ReadTree();
NUnit.Framework.Assert.IsTrue(tree != null);
NUnit.Framework.Assert.AreEqual(expected[i], tree.ToString());
}
Tree tree_1 = treeReader.ReadTree();
NUnit.Framework.Assert.IsFalse(tree_1 != null);
}
// We don't use valueOf because we sometimes use trees such as
// (bar (foo (foo 1))), and the default valueOf uses a
// TreeNormalizer that removes nodes from such a tree
public static Tree TreeFromString(string s)
{
try
{
ITreeReader tr = new PennTreeReader(new StringReader(s), new LabeledScoredTreeFactory());
return(tr.ReadTree());
}
catch (IOException e)
{
throw new Exception(e);
}
}
/// <summary>Process all the trees in the given directory.</summary>
/// <remarks>Process all the trees in the given directory. For example, the WSJ section of the Penn Treebank.</remarks>
/// <param name="name">The name of the directory we are processing.</param>
/// <param name="directory">The directory we are processing.</param>
/// <returns>
/// A dataset of subject/object pairs in the trees in the directory.
/// This is a list of sentences, such that each sentence has a collection of pairs of spans.
/// Each pair of spans is a subject/object span pair that constitutes a valid extraction.
/// </returns>
/// <exception cref="System.IO.IOException"/>
private static IList <Pair <ICoreMap, ICollection <Pair <Span, Span> > > > ProcessDirectory(string name, File directory)
{
Redwood.Util.ForceTrack("Processing " + name);
// Prepare the files to iterate over
IEnumerable <File> files = IOUtils.IterFilesRecursive(directory, "mrg");
int numTreesProcessed = 0;
IList <Pair <ICoreMap, ICollection <Pair <Span, Span> > > > trainingData = new List <Pair <ICoreMap, ICollection <Pair <Span, Span> > > >(1024);
// Iterate over the files
foreach (File file in files)
{
// log(file);
ITreeReader reader = new PennTreeReader(IOUtils.ReaderFromFile(file));
Tree tree;
while ((tree = reader.ReadTree()) != null)
{
try
{
// Prepare the tree
tree.IndexSpans();
tree.SetSpans();
// Get relevant information from sentence
IList <CoreLabel> tokens = tree.GetLeaves().Stream().Map(null).Collect(Collectors.ToList());
// .filter(leaf -> !TRACE_SOURCE_PATTERN.matcher(leaf.word()).matches() && !leaf.tag().equals("-NONE-"))
SemanticGraph graph = Parse(tree);
IDictionary <int, Span> targets = FindTraceTargets(tree);
IDictionary <int, int> sources = FindTraceSources(tree);
// Create a sentence object
ICoreMap sentence = new _ArrayCoreMap_325(tokens, graph, 4);
natlog.DoOneSentence(null, sentence);
// Generate training data
ICollection <Pair <Span, Span> > trainingDataFromSentence = SubjectObjectPairs(graph, tokens, targets, sources);
trainingData.Add(Pair.MakePair(sentence, trainingDataFromSentence));
// Debug print
numTreesProcessed += 1;
if (numTreesProcessed % 100 == 0)
{
Redwood.Util.Log("[" + new DecimalFormat("00000").Format(numTreesProcessed) + "] " + CountDatums(trainingData) + " known extractions");
}
}
catch (Exception t)
{
Sharpen.Runtime.PrintStackTrace(t);
}
}
}
// End
Redwood.Util.Log(string.Empty + numTreesProcessed + " trees processed yielding " + CountDatums(trainingData) + " known extractions");
Redwood.Util.EndTrack("Processing " + name);
return(trainingData);
}
