This package includes JAVA implementation of (Socher et al., NIPS 2011), and intermediate vector extraction of (Socher et al., ACL 2013).
Currently, this contains just trained results of the paper, but we planned to add other implemetations, such as:
- Stacked Denoised Auto Encoder
- Recursive Auto Encoder in (Socher et al., NIPS 2011)
- Unfolding Recursive Auto Encoder in (Socher et al., NIPS 2011)
Because jar with dependency file is not uploaded, you need to run maven install first.
If not, before this package is published in Maven repository, you need to specify dependencies of this package, which are declared in pom.xml.
To see how it works, you can simply run the jar file, with following shell script on /NeuralPhraseEmbedding directory:
maven clean install
java -jar ./NPE-latest-jar-with-dependencies.jarThen it automatically launches StanfordWrapper class. All you need to do is type the phrase when Phrase? prompt is shown.
This package contains javadoc jar file, NPE-latest-javadoc.jar.
Here is the sample code of StanfordWrapper main method
public static void main(String[] args) {
final StanfordWrapper instance = StanfordWrapper.getInstance();
final Scanner scan = new Scanner(System.in);
String line;
do {
System.out.print("Phrase? ");
line = scan.nextLine();
// Generate Parse Tree from Stanford RNN Parser.
final Tree tree = instance.parseTree(line);
// Get Phrase Vector(URAE)
final SimpleMatrix matrix = instance.getPhraseVectorOf(tree);
// If you want to get Syntactic Phrase Vector,
// final SimpleMatrix matrix = instance.getSyntacticScoreVectorOf(tree);
// If you want to get Syntactic Weight Matrix of that,
// final SimpleMatrix matrix = instance.getSyntacticWeightMatrixOf(tree);
System.out.println(matrix.toString());
System.out.println(tree.toString());
} while (line.length() > 0);
scan.close();
}