Examples with TreeTransformer edu.stanford.nlp.trees.TreeTransformer used on opensource projects

Example 6 with TreeTransformer

use of edu.stanford.nlp.trees.TreeTransformer in project CoreNLP by stanfordnlp.

the class SplitCanditoTrees method outputSplits.

/**
   * Right now this outputs trees in PTB format.  It outputs one tree
   * at a time until we have output enough trees to fill the given
   * file, then moves on to the next file.  Trees are output in the
   * order given in the <code>ids</code> list.
   * <br>
   * Trees have their words replaced with the words' lemmas, if those
   * lemmas exist.
   */
public static void outputSplits(List<String> ids, Map<String, Tree> treeMap) throws IOException {
    Queue<Integer> fSizeQueue = new LinkedList<>(Arrays.asList(fSizes));
    Queue<String> fNameQueue = new LinkedList<>(Arrays.asList(fNames));
    TregexPattern pBadTree = TregexPattern.compile("@SENT <: @PUNC");
    TregexPattern pBadTree2 = TregexPattern.compile("@SENT <1 @PUNC <2 @PUNC !<3 __");
    final TreeTransformer tt = new FTBCorrector();
    int size = fSizeQueue.remove();
    String filename = fNameQueue.remove();
    log.info("Outputing " + filename);
    PrintWriter writer = new PrintWriter(new BufferedWriter(new OutputStreamWriter(new FileOutputStream(filename), "UTF-8")));
    int outputCount = 0;
    for (String id : ids) {
        if (!treeMap.containsKey(id)) {
            log.info("Missing id: " + id);
            continue;
        }
        Tree tree = treeMap.get(id);
        TregexMatcher m = pBadTree.matcher(tree);
        TregexMatcher m2 = pBadTree2.matcher(tree);
        if (m.find() || m2.find()) {
            log.info("Discarding tree: " + tree.toString());
            continue;
        }
        // Punctuation normalization, etc.
        Tree backupCopy = tree.deepCopy();
        tree = tt.transformTree(tree);
        if (tree.firstChild().children().length == 0) {
            // Some trees have only punctuation. Tregex will mangle these. Don't throw those away.
            log.info("Saving tree: " + tree.toString());
            log.info("Backup: " + backupCopy.toString());
            tree = backupCopy;
        }
        if (LEMMAS_AS_LEAVES || ADD_MORPHO_TO_LEAVES) {
            mungeLeaves(tree, LEMMAS_AS_LEAVES, ADD_MORPHO_TO_LEAVES);
        }
        if (CC_TAGSET) {
            replacePOSTags(tree);
        }
        if (MORFETTE_OUTPUT) {
            writer.println(treeToMorfette(tree));
        } else {
            writer.println(tree.toString());
        }
        ++outputCount;
        if (outputCount == size) {
            outputCount = 0;
            size = fSizeQueue.remove();
            filename = fNameQueue.remove();
            log.info("Outputing " + filename);
            writer.close();
            writer = new PrintWriter(new BufferedWriter(new OutputStreamWriter(new FileOutputStream(filename), "UTF-8")));
        }
    }
    writer.close();
}

Example 7 with TreeTransformer

use of edu.stanford.nlp.trees.TreeTransformer in project CoreNLP by stanfordnlp.

the class FactoredParser method main.

/* some documentation for Roger's convenience
 * {pcfg,dep,combo}{PE,DE,TE} are precision/dep/tagging evals for the models

 * parser is the PCFG parser
 * dparser is the dependency parser
 * bparser is the combining parser

 * during testing:
 * tree is the test tree (gold tree)
 * binaryTree is the gold tree binarized
 * tree2b is the best PCFG paser, binarized
 * tree2 is the best PCFG parse (debinarized)
 * tree3 is the dependency parse, binarized
 * tree3db is the dependency parser, debinarized
 * tree4 is the best combo parse, binarized and then debinarized
 * tree4b is the best combo parse, binarized
 */
public static void main(String[] args) {
    Options op = new Options(new EnglishTreebankParserParams());
    // op.tlpParams may be changed to something else later, so don't use it till
    // after options are parsed.
    StringUtils.logInvocationString(log, args);
    String path = "/u/nlp/stuff/corpora/Treebank3/parsed/mrg/wsj";
    int trainLow = 200, trainHigh = 2199, testLow = 2200, testHigh = 2219;
    String serializeFile = null;
    int i = 0;
    while (i < args.length && args[i].startsWith("-")) {
        if (args[i].equalsIgnoreCase("-path") && (i + 1 < args.length)) {
            path = args[i + 1];
            i += 2;
        } else if (args[i].equalsIgnoreCase("-train") && (i + 2 < args.length)) {
            trainLow = Integer.parseInt(args[i + 1]);
            trainHigh = Integer.parseInt(args[i + 2]);
            i += 3;
        } else if (args[i].equalsIgnoreCase("-test") && (i + 2 < args.length)) {
            testLow = Integer.parseInt(args[i + 1]);
            testHigh = Integer.parseInt(args[i + 2]);
            i += 3;
        } else if (args[i].equalsIgnoreCase("-serialize") && (i + 1 < args.length)) {
            serializeFile = args[i + 1];
            i += 2;
        } else if (args[i].equalsIgnoreCase("-tLPP") && (i + 1 < args.length)) {
            try {
                op.tlpParams = (TreebankLangParserParams) Class.forName(args[i + 1]).newInstance();
            } catch (ClassNotFoundException e) {
                log.info("Class not found: " + args[i + 1]);
                throw new RuntimeException(e);
            } catch (InstantiationException e) {
                log.info("Couldn't instantiate: " + args[i + 1] + ": " + e.toString());
                throw new RuntimeException(e);
            } catch (IllegalAccessException e) {
                log.info("illegal access" + e);
                throw new RuntimeException(e);
            }
            i += 2;
        } else if (args[i].equals("-encoding")) {
            // sets encoding for TreebankLangParserParams
            op.tlpParams.setInputEncoding(args[i + 1]);
            op.tlpParams.setOutputEncoding(args[i + 1]);
            i += 2;
        } else {
            i = op.setOptionOrWarn(args, i);
        }
    }
    // System.out.println(tlpParams.getClass());
    TreebankLanguagePack tlp = op.tlpParams.treebankLanguagePack();
    op.trainOptions.sisterSplitters = Generics.newHashSet(Arrays.asList(op.tlpParams.sisterSplitters()));
    //    BinarizerFactory.TreeAnnotator.setTreebankLang(tlpParams);
    PrintWriter pw = op.tlpParams.pw();
    op.testOptions.display();
    op.trainOptions.display();
    op.display();
    op.tlpParams.display();
    // setup tree transforms
    Treebank trainTreebank = op.tlpParams.memoryTreebank();
    MemoryTreebank testTreebank = op.tlpParams.testMemoryTreebank();
    // Treebank blippTreebank = ((EnglishTreebankParserParams) tlpParams).diskTreebank();
    // String blippPath = "/afs/ir.stanford.edu/data/linguistic-data/BLLIP-WSJ/";
    // blippTreebank.loadPath(blippPath, "", true);
    Timing.startTime();
    log.info("Reading trees...");
    testTreebank.loadPath(path, new NumberRangeFileFilter(testLow, testHigh, true));
    if (op.testOptions.increasingLength) {
        Collections.sort(testTreebank, new TreeLengthComparator());
    }
    trainTreebank.loadPath(path, new NumberRangeFileFilter(trainLow, trainHigh, true));
    Timing.tick("done.");
    log.info("Binarizing trees...");
    TreeAnnotatorAndBinarizer binarizer;
    if (!op.trainOptions.leftToRight) {
        binarizer = new TreeAnnotatorAndBinarizer(op.tlpParams, op.forceCNF, !op.trainOptions.outsideFactor(), true, op);
    } else {
        binarizer = new TreeAnnotatorAndBinarizer(op.tlpParams.headFinder(), new LeftHeadFinder(), op.tlpParams, op.forceCNF, !op.trainOptions.outsideFactor(), true, op);
    }
    CollinsPuncTransformer collinsPuncTransformer = null;
    if (op.trainOptions.collinsPunc) {
        collinsPuncTransformer = new CollinsPuncTransformer(tlp);
    }
    TreeTransformer debinarizer = new Debinarizer(op.forceCNF);
    List<Tree> binaryTrainTrees = new ArrayList<>();
    if (op.trainOptions.selectiveSplit) {
        op.trainOptions.splitters = ParentAnnotationStats.getSplitCategories(trainTreebank, op.trainOptions.tagSelectiveSplit, 0, op.trainOptions.selectiveSplitCutOff, op.trainOptions.tagSelectiveSplitCutOff, op.tlpParams.treebankLanguagePack());
        if (op.trainOptions.deleteSplitters != null) {
            List<String> deleted = new ArrayList<>();
            for (String del : op.trainOptions.deleteSplitters) {
                String baseDel = tlp.basicCategory(del);
                boolean checkBasic = del.equals(baseDel);
                for (Iterator<String> it = op.trainOptions.splitters.iterator(); it.hasNext(); ) {
                    String elem = it.next();
                    String baseElem = tlp.basicCategory(elem);
                    boolean delStr = checkBasic && baseElem.equals(baseDel) || elem.equals(del);
                    if (delStr) {
                        it.remove();
                        deleted.add(elem);
                    }
                }
            }
            log.info("Removed from vertical splitters: " + deleted);
        }
    }
    if (op.trainOptions.selectivePostSplit) {
        TreeTransformer myTransformer = new TreeAnnotator(op.tlpParams.headFinder(), op.tlpParams, op);
        Treebank annotatedTB = trainTreebank.transform(myTransformer);
        op.trainOptions.postSplitters = ParentAnnotationStats.getSplitCategories(annotatedTB, true, 0, op.trainOptions.selectivePostSplitCutOff, op.trainOptions.tagSelectivePostSplitCutOff, op.tlpParams.treebankLanguagePack());
    }
    if (op.trainOptions.hSelSplit) {
        binarizer.setDoSelectiveSplit(false);
        for (Tree tree : trainTreebank) {
            if (op.trainOptions.collinsPunc) {
                tree = collinsPuncTransformer.transformTree(tree);
            }
            //tree.pennPrint(tlpParams.pw());
            tree = binarizer.transformTree(tree);
        //binaryTrainTrees.add(tree);
        }
        binarizer.setDoSelectiveSplit(true);
    }
    for (Tree tree : trainTreebank) {
        if (op.trainOptions.collinsPunc) {
            tree = collinsPuncTransformer.transformTree(tree);
        }
        tree = binarizer.transformTree(tree);
        binaryTrainTrees.add(tree);
    }
    if (op.testOptions.verbose) {
        binarizer.dumpStats();
    }
    List<Tree> binaryTestTrees = new ArrayList<>();
    for (Tree tree : testTreebank) {
        if (op.trainOptions.collinsPunc) {
            tree = collinsPuncTransformer.transformTree(tree);
        }
        tree = binarizer.transformTree(tree);
        binaryTestTrees.add(tree);
    }
    // binarization
    Timing.tick("done.");
    BinaryGrammar bg = null;
    UnaryGrammar ug = null;
    DependencyGrammar dg = null;
    // DependencyGrammar dgBLIPP = null;
    Lexicon lex = null;
    Index<String> stateIndex = new HashIndex<>();
    // extract grammars
    Extractor<Pair<UnaryGrammar, BinaryGrammar>> bgExtractor = new BinaryGrammarExtractor(op, stateIndex);
    if (op.doPCFG) {
        log.info("Extracting PCFG...");
        Pair<UnaryGrammar, BinaryGrammar> bgug = null;
        if (op.trainOptions.cheatPCFG) {
            List<Tree> allTrees = new ArrayList<>(binaryTrainTrees);
            allTrees.addAll(binaryTestTrees);
            bgug = bgExtractor.extract(allTrees);
        } else {
            bgug = bgExtractor.extract(binaryTrainTrees);
        }
        bg = bgug.second;
        bg.splitRules();
        ug = bgug.first;
        ug.purgeRules();
        Timing.tick("done.");
    }
    log.info("Extracting Lexicon...");
    Index<String> wordIndex = new HashIndex<>();
    Index<String> tagIndex = new HashIndex<>();
    lex = op.tlpParams.lex(op, wordIndex, tagIndex);
    lex.initializeTraining(binaryTrainTrees.size());
    lex.train(binaryTrainTrees);
    lex.finishTraining();
    Timing.tick("done.");
    if (op.doDep) {
        log.info("Extracting Dependencies...");
        binaryTrainTrees.clear();
        Extractor<DependencyGrammar> dgExtractor = new MLEDependencyGrammarExtractor(op, wordIndex, tagIndex);
        // dgBLIPP = (DependencyGrammar) dgExtractor.extract(new ConcatenationIterator(trainTreebank.iterator(),blippTreebank.iterator()),new TransformTreeDependency(tlpParams,true));
        // DependencyGrammar dg1 = dgExtractor.extract(trainTreebank.iterator(), new TransformTreeDependency(op.tlpParams, true));
        //dgBLIPP=(DependencyGrammar)dgExtractor.extract(blippTreebank.iterator(),new TransformTreeDependency(tlpParams));
        //dg = (DependencyGrammar) dgExtractor.extract(new ConcatenationIterator(trainTreebank.iterator(),blippTreebank.iterator()),new TransformTreeDependency(tlpParams));
        // dg=new DependencyGrammarCombination(dg1,dgBLIPP,2);
        //uses information whether the words are known or not, discards unknown words
        dg = dgExtractor.extract(binaryTrainTrees);
        Timing.tick("done.");
        //System.out.print("Extracting Unknown Word Model...");
        //UnknownWordModel uwm = (UnknownWordModel)uwmExtractor.extract(binaryTrainTrees);
        //Timing.tick("done.");
        System.out.print("Tuning Dependency Model...");
        dg.tune(binaryTestTrees);
        //System.out.println("TUNE DEPS: "+tuneDeps);
        Timing.tick("done.");
    }
    BinaryGrammar boundBG = bg;
    UnaryGrammar boundUG = ug;
    GrammarProjection gp = new NullGrammarProjection(bg, ug);
    // serialization
    if (serializeFile != null) {
        log.info("Serializing parser...");
        LexicalizedParser parser = new LexicalizedParser(lex, bg, ug, dg, stateIndex, wordIndex, tagIndex, op);
        parser.saveParserToSerialized(serializeFile);
        Timing.tick("done.");
    }
    // test: pcfg-parse and output
    ExhaustivePCFGParser parser = null;
    if (op.doPCFG) {
        parser = new ExhaustivePCFGParser(boundBG, boundUG, lex, op, stateIndex, wordIndex, tagIndex);
    }
    ExhaustiveDependencyParser dparser = ((op.doDep && !op.testOptions.useFastFactored) ? new ExhaustiveDependencyParser(dg, lex, op, wordIndex, tagIndex) : null);
    Scorer scorer = (op.doPCFG ? new TwinScorer(new ProjectionScorer(parser, gp, op), dparser) : null);
    //Scorer scorer = parser;
    BiLexPCFGParser bparser = null;
    if (op.doPCFG && op.doDep) {
        bparser = (op.testOptions.useN5) ? new BiLexPCFGParser.N5BiLexPCFGParser(scorer, parser, dparser, bg, ug, dg, lex, op, gp, stateIndex, wordIndex, tagIndex) : new BiLexPCFGParser(scorer, parser, dparser, bg, ug, dg, lex, op, gp, stateIndex, wordIndex, tagIndex);
    }
    Evalb pcfgPE = new Evalb("pcfg  PE", true);
    Evalb comboPE = new Evalb("combo PE", true);
    AbstractEval pcfgCB = new Evalb.CBEval("pcfg  CB", true);
    AbstractEval pcfgTE = new TaggingEval("pcfg  TE");
    AbstractEval comboTE = new TaggingEval("combo TE");
    AbstractEval pcfgTEnoPunct = new TaggingEval("pcfg nopunct TE");
    AbstractEval comboTEnoPunct = new TaggingEval("combo nopunct TE");
    AbstractEval depTE = new TaggingEval("depnd TE");
    AbstractEval depDE = new UnlabeledAttachmentEval("depnd DE", true, null, tlp.punctuationWordRejectFilter());
    AbstractEval comboDE = new UnlabeledAttachmentEval("combo DE", true, null, tlp.punctuationWordRejectFilter());
    if (op.testOptions.evalb) {
        EvalbFormatWriter.initEVALBfiles(op.tlpParams);
    }
    // int[] countByLength = new int[op.testOptions.maxLength+1];
    // Use a reflection ruse, so one can run this without needing the
    // tagger.  Using a function rather than a MaxentTagger means we
    // can distribute a version of the parser that doesn't include the
    // entire tagger.
    Function<List<? extends HasWord>, ArrayList<TaggedWord>> tagger = null;
    if (op.testOptions.preTag) {
        try {
            Class[] argsClass = { String.class };
            Object[] arguments = new Object[] { op.testOptions.taggerSerializedFile };
            tagger = (Function<List<? extends HasWord>, ArrayList<TaggedWord>>) Class.forName("edu.stanford.nlp.tagger.maxent.MaxentTagger").getConstructor(argsClass).newInstance(arguments);
        } catch (Exception e) {
            log.info(e);
            log.info("Warning: No pretagging of sentences will be done.");
        }
    }
    for (int tNum = 0, ttSize = testTreebank.size(); tNum < ttSize; tNum++) {
        Tree tree = testTreebank.get(tNum);
        int testTreeLen = tree.yield().size();
        if (testTreeLen > op.testOptions.maxLength) {
            continue;
        }
        Tree binaryTree = binaryTestTrees.get(tNum);
        // countByLength[testTreeLen]++;
        System.out.println("-------------------------------------");
        System.out.println("Number: " + (tNum + 1));
        System.out.println("Length: " + testTreeLen);
        //tree.pennPrint(pw);
        // System.out.println("XXXX The binary tree is");
        // binaryTree.pennPrint(pw);
        //System.out.println("Here are the tags in the lexicon:");
        //System.out.println(lex.showTags());
        //System.out.println("Here's the tagnumberer:");
        //System.out.println(Numberer.getGlobalNumberer("tags").toString());
        long timeMil1 = System.currentTimeMillis();
        Timing.tick("Starting parse.");
        if (op.doPCFG) {
            //log.info(op.testOptions.forceTags);
            if (op.testOptions.forceTags) {
                if (tagger != null) {
                    //System.out.println("Using a tagger to set tags");
                    //System.out.println("Tagged sentence as: " + tagger.processSentence(cutLast(wordify(binaryTree.yield()))).toString(false));
                    parser.parse(addLast(tagger.apply(cutLast(wordify(binaryTree.yield())))));
                } else {
                    //System.out.println("Forcing tags to match input.");
                    parser.parse(cleanTags(binaryTree.taggedYield(), tlp));
                }
            } else {
                // System.out.println("XXXX Parsing " + binaryTree.yield());
                parser.parse(binaryTree.yieldHasWord());
            }
        //Timing.tick("Done with pcfg phase.");
        }
        if (op.doDep) {
            dparser.parse(binaryTree.yieldHasWord());
        //Timing.tick("Done with dependency phase.");
        }
        boolean bothPassed = false;
        if (op.doPCFG && op.doDep) {
            bothPassed = bparser.parse(binaryTree.yieldHasWord());
        //Timing.tick("Done with combination phase.");
        }
        long timeMil2 = System.currentTimeMillis();
        long elapsed = timeMil2 - timeMil1;
        log.info("Time: " + ((int) (elapsed / 100)) / 10.00 + " sec.");
        //System.out.println("PCFG Best Parse:");
        Tree tree2b = null;
        Tree tree2 = null;
        //System.out.println("Got full best parse...");
        if (op.doPCFG) {
            tree2b = parser.getBestParse();
            tree2 = debinarizer.transformTree(tree2b);
        }
        //System.out.println("Debinarized parse...");
        //tree2.pennPrint();
        //System.out.println("DepG Best Parse:");
        Tree tree3 = null;
        Tree tree3db = null;
        if (op.doDep) {
            tree3 = dparser.getBestParse();
            // was: but wrong Tree tree3db = debinarizer.transformTree(tree2);
            tree3db = debinarizer.transformTree(tree3);
            tree3.pennPrint(pw);
        }
        //tree.pennPrint();
        //((Tree)binaryTrainTrees.get(tNum)).pennPrint();
        //System.out.println("Combo Best Parse:");
        Tree tree4 = null;
        if (op.doPCFG && op.doDep) {
            try {
                tree4 = bparser.getBestParse();
                if (tree4 == null) {
                    tree4 = tree2b;
                }
            } catch (NullPointerException e) {
                log.info("Blocked, using PCFG parse!");
                tree4 = tree2b;
            }
        }
        if (op.doPCFG && !bothPassed) {
            tree4 = tree2b;
        }
        //tree4.pennPrint();
        if (op.doDep) {
            depDE.evaluate(tree3, binaryTree, pw);
            depTE.evaluate(tree3db, tree, pw);
        }
        TreeTransformer tc = op.tlpParams.collinizer();
        TreeTransformer tcEvalb = op.tlpParams.collinizerEvalb();
        if (op.doPCFG) {
            // System.out.println("XXXX Best PCFG was: ");
            // tree2.pennPrint();
            // System.out.println("XXXX Transformed best PCFG is: ");
            // tc.transformTree(tree2).pennPrint();
            //System.out.println("True Best Parse:");
            //tree.pennPrint();
            //tc.transformTree(tree).pennPrint();
            pcfgPE.evaluate(tc.transformTree(tree2), tc.transformTree(tree), pw);
            pcfgCB.evaluate(tc.transformTree(tree2), tc.transformTree(tree), pw);
            Tree tree4b = null;
            if (op.doDep) {
                comboDE.evaluate((bothPassed ? tree4 : tree3), binaryTree, pw);
                tree4b = tree4;
                tree4 = debinarizer.transformTree(tree4);
                if (op.nodePrune) {
                    NodePruner np = new NodePruner(parser, debinarizer);
                    tree4 = np.prune(tree4);
                }
                //tree4.pennPrint();
                comboPE.evaluate(tc.transformTree(tree4), tc.transformTree(tree), pw);
            }
            //pcfgTE.evaluate(tree2, tree);
            pcfgTE.evaluate(tcEvalb.transformTree(tree2), tcEvalb.transformTree(tree), pw);
            pcfgTEnoPunct.evaluate(tc.transformTree(tree2), tc.transformTree(tree), pw);
            if (op.doDep) {
                comboTE.evaluate(tcEvalb.transformTree(tree4), tcEvalb.transformTree(tree), pw);
                comboTEnoPunct.evaluate(tc.transformTree(tree4), tc.transformTree(tree), pw);
            }
            System.out.println("PCFG only: " + parser.scoreBinarizedTree(tree2b, 0));
            //tc.transformTree(tree2).pennPrint();
            tree2.pennPrint(pw);
            if (op.doDep) {
                System.out.println("Combo: " + parser.scoreBinarizedTree(tree4b, 0));
                // tc.transformTree(tree4).pennPrint(pw);
                tree4.pennPrint(pw);
            }
            System.out.println("Correct:" + parser.scoreBinarizedTree(binaryTree, 0));
            /*
        if (parser.scoreBinarizedTree(tree2b,true) < parser.scoreBinarizedTree(binaryTree,true)) {
          System.out.println("SCORE INVERSION");
          parser.validateBinarizedTree(binaryTree,0);
        }
        */
            tree.pennPrint(pw);
        }
        if (op.testOptions.evalb) {
            if (op.doPCFG && op.doDep) {
                EvalbFormatWriter.writeEVALBline(tcEvalb.transformTree(tree), tcEvalb.transformTree(tree4));
            } else if (op.doPCFG) {
                EvalbFormatWriter.writeEVALBline(tcEvalb.transformTree(tree), tcEvalb.transformTree(tree2));
            } else if (op.doDep) {
                EvalbFormatWriter.writeEVALBline(tcEvalb.transformTree(tree), tcEvalb.transformTree(tree3db));
            }
        }
    }
    if (op.testOptions.evalb) {
        EvalbFormatWriter.closeEVALBfiles();
    }
    // op.testOptions.display();
    if (op.doPCFG) {
        pcfgPE.display(false, pw);
        System.out.println("Grammar size: " + stateIndex.size());
        pcfgCB.display(false, pw);
        if (op.doDep) {
            comboPE.display(false, pw);
        }
        pcfgTE.display(false, pw);
        pcfgTEnoPunct.display(false, pw);
        if (op.doDep) {
            comboTE.display(false, pw);
            comboTEnoPunct.display(false, pw);
        }
    }
    if (op.doDep) {
        depTE.display(false, pw);
        depDE.display(false, pw);
    }
    if (op.doPCFG && op.doDep) {
        comboDE.display(false, pw);
    }
// pcfgPE.printGoodBad();
}

Example 8 with TreeTransformer

use of edu.stanford.nlp.trees.TreeTransformer in project CoreNLP by stanfordnlp.

the class ChineseCharacterBasedLexiconTraining method main.

public static void main(String[] args) throws IOException {
    Map<String, Integer> flagsToNumArgs = Generics.newHashMap();
    flagsToNumArgs.put("-parser", Integer.valueOf(3));
    flagsToNumArgs.put("-lex", Integer.valueOf(3));
    flagsToNumArgs.put("-test", Integer.valueOf(2));
    flagsToNumArgs.put("-out", Integer.valueOf(1));
    flagsToNumArgs.put("-lengthPenalty", Integer.valueOf(1));
    flagsToNumArgs.put("-penaltyType", Integer.valueOf(1));
    flagsToNumArgs.put("-maxLength", Integer.valueOf(1));
    flagsToNumArgs.put("-stats", Integer.valueOf(2));
    Map<String, String[]> argMap = StringUtils.argsToMap(args, flagsToNumArgs);
    boolean eval = argMap.containsKey("-eval");
    PrintWriter pw = null;
    if (argMap.containsKey("-out")) {
        pw = new PrintWriter(new OutputStreamWriter(new FileOutputStream((argMap.get("-out"))[0]), "GB18030"), true);
    }
    log.info("ChineseCharacterBasedLexicon called with args:");
    ChineseTreebankParserParams ctpp = new ChineseTreebankParserParams();
    for (int i = 0; i < args.length; i++) {
        ctpp.setOptionFlag(args, i);
        log.info(" " + args[i]);
    }
    log.info();
    Options op = new Options(ctpp);
    if (argMap.containsKey("-stats")) {
        String[] statArgs = (argMap.get("-stats"));
        MemoryTreebank rawTrainTreebank = op.tlpParams.memoryTreebank();
        FileFilter trainFilt = new NumberRangesFileFilter(statArgs[1], false);
        rawTrainTreebank.loadPath(new File(statArgs[0]), trainFilt);
        log.info("Done reading trees.");
        MemoryTreebank trainTreebank;
        if (argMap.containsKey("-annotate")) {
            trainTreebank = new MemoryTreebank();
            TreeAnnotator annotator = new TreeAnnotator(ctpp.headFinder(), ctpp, op);
            for (Tree tree : rawTrainTreebank) {
                trainTreebank.add(annotator.transformTree(tree));
            }
            log.info("Done annotating trees.");
        } else {
            trainTreebank = rawTrainTreebank;
        }
        printStats(trainTreebank, pw);
        System.exit(0);
    }
    int maxLength = 1000000;
    //    Test.verbose = true;
    if (argMap.containsKey("-norm")) {
        op.testOptions.lengthNormalization = true;
    }
    if (argMap.containsKey("-maxLength")) {
        maxLength = Integer.parseInt((argMap.get("-maxLength"))[0]);
    }
    op.testOptions.maxLength = 120;
    boolean combo = argMap.containsKey("-combo");
    if (combo) {
        ctpp.useCharacterBasedLexicon = true;
        op.testOptions.maxSpanForTags = 10;
        op.doDep = false;
        op.dcTags = false;
    }
    LexicalizedParser lp = null;
    Lexicon lex = null;
    if (argMap.containsKey("-parser")) {
        String[] parserArgs = (argMap.get("-parser"));
        if (parserArgs.length > 1) {
            FileFilter trainFilt = new NumberRangesFileFilter(parserArgs[1], false);
            lp = LexicalizedParser.trainFromTreebank(parserArgs[0], trainFilt, op);
            if (parserArgs.length == 3) {
                String filename = parserArgs[2];
                log.info("Writing parser in serialized format to file " + filename + " ");
                System.err.flush();
                ObjectOutputStream out = IOUtils.writeStreamFromString(filename);
                out.writeObject(lp);
                out.close();
                log.info("done.");
            }
        } else {
            String parserFile = parserArgs[0];
            lp = LexicalizedParser.loadModel(parserFile, op);
        }
        lex = lp.getLexicon();
        op = lp.getOp();
        ctpp = (ChineseTreebankParserParams) op.tlpParams;
    }
    if (argMap.containsKey("-rad")) {
        ctpp.useUnknownCharacterModel = true;
    }
    if (argMap.containsKey("-lengthPenalty")) {
        ctpp.lengthPenalty = Double.parseDouble((argMap.get("-lengthPenalty"))[0]);
    }
    if (argMap.containsKey("-penaltyType")) {
        ctpp.penaltyType = Integer.parseInt((argMap.get("-penaltyType"))[0]);
    }
    if (argMap.containsKey("-lex")) {
        String[] lexArgs = (argMap.get("-lex"));
        if (lexArgs.length > 1) {
            Index<String> wordIndex = new HashIndex<>();
            Index<String> tagIndex = new HashIndex<>();
            lex = ctpp.lex(op, wordIndex, tagIndex);
            MemoryTreebank rawTrainTreebank = op.tlpParams.memoryTreebank();
            FileFilter trainFilt = new NumberRangesFileFilter(lexArgs[1], false);
            rawTrainTreebank.loadPath(new File(lexArgs[0]), trainFilt);
            log.info("Done reading trees.");
            MemoryTreebank trainTreebank;
            if (argMap.containsKey("-annotate")) {
                trainTreebank = new MemoryTreebank();
                TreeAnnotator annotator = new TreeAnnotator(ctpp.headFinder(), ctpp, op);
                for (Tree tree : rawTrainTreebank) {
                    tree = annotator.transformTree(tree);
                    trainTreebank.add(tree);
                }
                log.info("Done annotating trees.");
            } else {
                trainTreebank = rawTrainTreebank;
            }
            lex.initializeTraining(trainTreebank.size());
            lex.train(trainTreebank);
            lex.finishTraining();
            log.info("Done training lexicon.");
            if (lexArgs.length == 3) {
                String filename = lexArgs.length == 3 ? lexArgs[2] : "parsers/chineseCharLex.ser.gz";
                log.info("Writing lexicon in serialized format to file " + filename + " ");
                System.err.flush();
                ObjectOutputStream out = IOUtils.writeStreamFromString(filename);
                out.writeObject(lex);
                out.close();
                log.info("done.");
            }
        } else {
            String lexFile = lexArgs.length == 1 ? lexArgs[0] : "parsers/chineseCharLex.ser.gz";
            log.info("Reading Lexicon from file " + lexFile);
            ObjectInputStream in = IOUtils.readStreamFromString(lexFile);
            try {
                lex = (Lexicon) in.readObject();
            } catch (ClassNotFoundException e) {
                throw new RuntimeException("Bad serialized file: " + lexFile);
            }
            in.close();
        }
    }
    if (argMap.containsKey("-test")) {
        boolean segmentWords = ctpp.segment;
        boolean parse = lp != null;
        assert (parse || segmentWords);
        //      WordCatConstituent.collinizeWords = argMap.containsKey("-collinizeWords");
        //      WordCatConstituent.collinizeTags = argMap.containsKey("-collinizeTags");
        WordSegmenter seg = null;
        if (segmentWords) {
            seg = (WordSegmenter) lex;
        }
        String[] testArgs = (argMap.get("-test"));
        MemoryTreebank testTreebank = op.tlpParams.memoryTreebank();
        FileFilter testFilt = new NumberRangesFileFilter(testArgs[1], false);
        testTreebank.loadPath(new File(testArgs[0]), testFilt);
        TreeTransformer subcategoryStripper = op.tlpParams.subcategoryStripper();
        TreeTransformer collinizer = ctpp.collinizer();
        WordCatEquivalenceClasser eqclass = new WordCatEquivalenceClasser();
        WordCatEqualityChecker eqcheck = new WordCatEqualityChecker();
        EquivalenceClassEval basicEval = new EquivalenceClassEval(eqclass, eqcheck, "basic");
        EquivalenceClassEval collinsEval = new EquivalenceClassEval(eqclass, eqcheck, "collinized");
        List<String> evalTypes = new ArrayList<>(3);
        boolean goodPOS = false;
        if (segmentWords) {
            evalTypes.add(WordCatConstituent.wordType);
            if (ctpp.segmentMarkov && !parse) {
                evalTypes.add(WordCatConstituent.tagType);
                goodPOS = true;
            }
        }
        if (parse) {
            evalTypes.add(WordCatConstituent.tagType);
            evalTypes.add(WordCatConstituent.catType);
            if (combo) {
                evalTypes.add(WordCatConstituent.wordType);
                goodPOS = true;
            }
        }
        TreeToBracketProcessor proc = new TreeToBracketProcessor(evalTypes);
        log.info("Testing...");
        for (Tree goldTop : testTreebank) {
            Tree gold = goldTop.firstChild();
            List<HasWord> goldSentence = gold.yieldHasWord();
            if (goldSentence.size() > maxLength) {
                log.info("Skipping sentence; too long: " + goldSentence.size());
                continue;
            } else {
                log.info("Processing sentence; length: " + goldSentence.size());
            }
            List<HasWord> s;
            if (segmentWords) {
                StringBuilder goldCharBuf = new StringBuilder();
                for (HasWord aGoldSentence : goldSentence) {
                    StringLabel word = (StringLabel) aGoldSentence;
                    goldCharBuf.append(word.value());
                }
                String goldChars = goldCharBuf.toString();
                s = seg.segment(goldChars);
            } else {
                s = goldSentence;
            }
            Tree tree;
            if (parse) {
                tree = lp.parseTree(s);
                if (tree == null) {
                    throw new RuntimeException("PARSER RETURNED NULL!!!");
                }
            } else {
                tree = Trees.toFlatTree(s);
                tree = subcategoryStripper.transformTree(tree);
            }
            if (pw != null) {
                if (parse) {
                    tree.pennPrint(pw);
                } else {
                    Iterator sentIter = s.iterator();
                    for (; ; ) {
                        Word word = (Word) sentIter.next();
                        pw.print(word.word());
                        if (sentIter.hasNext()) {
                            pw.print(" ");
                        } else {
                            break;
                        }
                    }
                }
                pw.println();
            }
            if (eval) {
                Collection ourBrackets, goldBrackets;
                ourBrackets = proc.allBrackets(tree);
                goldBrackets = proc.allBrackets(gold);
                if (goodPOS) {
                    ourBrackets.addAll(proc.commonWordTagTypeBrackets(tree, gold));
                    goldBrackets.addAll(proc.commonWordTagTypeBrackets(gold, tree));
                }
                basicEval.eval(ourBrackets, goldBrackets);
                System.out.println("\nScores:");
                basicEval.displayLast();
                Tree collinsTree = collinizer.transformTree(tree);
                Tree collinsGold = collinizer.transformTree(gold);
                ourBrackets = proc.allBrackets(collinsTree);
                goldBrackets = proc.allBrackets(collinsGold);
                if (goodPOS) {
                    ourBrackets.addAll(proc.commonWordTagTypeBrackets(collinsTree, collinsGold));
                    goldBrackets.addAll(proc.commonWordTagTypeBrackets(collinsGold, collinsTree));
                }
                collinsEval.eval(ourBrackets, goldBrackets);
                System.out.println("\nCollinized scores:");
                collinsEval.displayLast();
                System.out.println();
            }
        }
        if (eval) {
            basicEval.display();
            System.out.println();
            collinsEval.display();
        }
    }
}

Example 9 with TreeTransformer

use of edu.stanford.nlp.trees.TreeTransformer in project CoreNLP by stanfordnlp.

the class DVParser method main.

/**
   * An example command line for training a new parser:
   * <br>
   *  nohup java -mx6g edu.stanford.nlp.parser.dvparser.DVParser -cachedTrees /scr/nlp/data/dvparser/wsj/cached.wsj.train.simple.ser.gz -train -testTreebank  /afs/ir/data/linguistic-data/Treebank/3/parsed/mrg/wsj/22 2200-2219 -debugOutputFrequency 400 -nofilter -trainingThreads 5 -parser /u/nlp/data/lexparser/wsjPCFG.nocompact.simple.ser.gz -trainingIterations 40 -batchSize 25 -model /scr/nlp/data/dvparser/wsj/wsj.combine.v2.ser.gz -unkWord "*UNK*" -dvCombineCategories &gt; /scr/nlp/data/dvparser/wsj/wsj.combine.v2.out 2&gt;&amp;1 &amp;
   */
public static void main(String[] args) throws IOException, ClassNotFoundException {
    if (args.length == 0) {
        help();
        System.exit(2);
    }
    log.info("Running DVParser with arguments:");
    for (String arg : args) {
        log.info("  " + arg);
    }
    log.info();
    String parserPath = null;
    String trainTreebankPath = null;
    FileFilter trainTreebankFilter = null;
    String cachedTrainTreesPath = null;
    boolean runGradientCheck = false;
    boolean runTraining = false;
    String testTreebankPath = null;
    FileFilter testTreebankFilter = null;
    String initialModelPath = null;
    String modelPath = null;
    boolean filter = true;
    String resultsRecordPath = null;
    List<String> unusedArgs = new ArrayList<>();
    // These parameters can be null or 0 if the model was not
    // serialized with the new parameters.  Setting the options at the
    // command line will override these defaults.
    // TODO: if/when we integrate back into the main branch and
    // rebuild models, we can get rid of this
    List<String> argsWithDefaults = new ArrayList<>(Arrays.asList(new String[] { "-wordVectorFile", Options.LexOptions.DEFAULT_WORD_VECTOR_FILE, "-dvKBest", Integer.toString(TrainOptions.DEFAULT_K_BEST), "-batchSize", Integer.toString(TrainOptions.DEFAULT_BATCH_SIZE), "-trainingIterations", Integer.toString(TrainOptions.DEFAULT_TRAINING_ITERATIONS), "-qnIterationsPerBatch", Integer.toString(TrainOptions.DEFAULT_QN_ITERATIONS_PER_BATCH), "-regCost", Double.toString(TrainOptions.DEFAULT_REGCOST), "-learningRate", Double.toString(TrainOptions.DEFAULT_LEARNING_RATE), "-deltaMargin", Double.toString(TrainOptions.DEFAULT_DELTA_MARGIN), "-unknownNumberVector", "-unknownDashedWordVectors", "-unknownCapsVector", "-unknownchinesepercentvector", "-unknownchinesenumbervector", "-unknownchineseyearvector", "-unkWord", "*UNK*", "-transformMatrixType", "DIAGONAL", "-scalingForInit", Double.toString(TrainOptions.DEFAULT_SCALING_FOR_INIT), "-trainWordVectors" }));
    argsWithDefaults.addAll(Arrays.asList(args));
    args = argsWithDefaults.toArray(new String[argsWithDefaults.size()]);
    for (int argIndex = 0; argIndex < args.length; ) {
        if (args[argIndex].equalsIgnoreCase("-parser")) {
            parserPath = args[argIndex + 1];
            argIndex += 2;
        } else if (args[argIndex].equalsIgnoreCase("-testTreebank")) {
            Pair<String, FileFilter> treebankDescription = ArgUtils.getTreebankDescription(args, argIndex, "-testTreebank");
            argIndex = argIndex + ArgUtils.numSubArgs(args, argIndex) + 1;
            testTreebankPath = treebankDescription.first();
            testTreebankFilter = treebankDescription.second();
        } else if (args[argIndex].equalsIgnoreCase("-treebank")) {
            Pair<String, FileFilter> treebankDescription = ArgUtils.getTreebankDescription(args, argIndex, "-treebank");
            argIndex = argIndex + ArgUtils.numSubArgs(args, argIndex) + 1;
            trainTreebankPath = treebankDescription.first();
            trainTreebankFilter = treebankDescription.second();
        } else if (args[argIndex].equalsIgnoreCase("-cachedTrees")) {
            cachedTrainTreesPath = args[argIndex + 1];
            argIndex += 2;
        } else if (args[argIndex].equalsIgnoreCase("-runGradientCheck")) {
            runGradientCheck = true;
            argIndex++;
        } else if (args[argIndex].equalsIgnoreCase("-train")) {
            runTraining = true;
            argIndex++;
        } else if (args[argIndex].equalsIgnoreCase("-model")) {
            modelPath = args[argIndex + 1];
            argIndex += 2;
        } else if (args[argIndex].equalsIgnoreCase("-nofilter")) {
            filter = false;
            argIndex++;
        } else if (args[argIndex].equalsIgnoreCase("-continueTraining")) {
            runTraining = true;
            filter = false;
            initialModelPath = args[argIndex + 1];
            argIndex += 2;
        } else if (args[argIndex].equalsIgnoreCase("-resultsRecord")) {
            resultsRecordPath = args[argIndex + 1];
            argIndex += 2;
        } else {
            unusedArgs.add(args[argIndex++]);
        }
    }
    if (parserPath == null && modelPath == null) {
        throw new IllegalArgumentException("Must supply either a base parser model with -parser or a serialized DVParser with -model");
    }
    if (!runTraining && modelPath == null && !runGradientCheck) {
        throw new IllegalArgumentException("Need to either train a new model, run the gradient check or specify a model to load with -model");
    }
    String[] newArgs = unusedArgs.toArray(new String[unusedArgs.size()]);
    DVParser dvparser = null;
    LexicalizedParser lexparser = null;
    if (initialModelPath != null) {
        lexparser = LexicalizedParser.loadModel(initialModelPath, newArgs);
        DVModel model = getModelFromLexicalizedParser(lexparser);
        dvparser = new DVParser(model, lexparser);
    } else if (runTraining || runGradientCheck) {
        lexparser = LexicalizedParser.loadModel(parserPath, newArgs);
        dvparser = new DVParser(lexparser);
    } else if (modelPath != null) {
        lexparser = LexicalizedParser.loadModel(modelPath, newArgs);
        DVModel model = getModelFromLexicalizedParser(lexparser);
        dvparser = new DVParser(model, lexparser);
    }
    List<Tree> trainSentences = new ArrayList<>();
    IdentityHashMap<Tree, byte[]> trainCompressedParses = Generics.newIdentityHashMap();
    if (cachedTrainTreesPath != null) {
        for (String path : cachedTrainTreesPath.split(",")) {
            List<Pair<Tree, byte[]>> cache = IOUtils.readObjectFromFile(path);
            for (Pair<Tree, byte[]> pair : cache) {
                trainSentences.add(pair.first());
                trainCompressedParses.put(pair.first(), pair.second());
            }
            log.info("Read in " + cache.size() + " trees from " + path);
        }
    }
    if (trainTreebankPath != null) {
        // TODO: make the transformer a member of the model?
        TreeTransformer transformer = buildTrainTransformer(dvparser.getOp());
        Treebank treebank = dvparser.getOp().tlpParams.memoryTreebank();
        ;
        treebank.loadPath(trainTreebankPath, trainTreebankFilter);
        treebank = treebank.transform(transformer);
        log.info("Read in " + treebank.size() + " trees from " + trainTreebankPath);
        CacheParseHypotheses cacher = new CacheParseHypotheses(dvparser.parser);
        CacheParseHypotheses.CacheProcessor processor = new CacheParseHypotheses.CacheProcessor(cacher, lexparser, dvparser.op.trainOptions.dvKBest, transformer);
        for (Tree tree : treebank) {
            trainSentences.add(tree);
            trainCompressedParses.put(tree, processor.process(tree).second);
        //System.out.println(tree);
        }
        log.info("Finished parsing " + treebank.size() + " trees, getting " + dvparser.op.trainOptions.dvKBest + " hypotheses each");
    }
    if ((runTraining || runGradientCheck) && filter) {
        log.info("Filtering rules for the given training set");
        dvparser.dvModel.setRulesForTrainingSet(trainSentences, trainCompressedParses);
        log.info("Done filtering rules; " + dvparser.dvModel.numBinaryMatrices + " binary matrices, " + dvparser.dvModel.numUnaryMatrices + " unary matrices, " + dvparser.dvModel.wordVectors.size() + " word vectors");
    }
    //dvparser.dvModel.printAllMatrices();
    Treebank testTreebank = null;
    if (testTreebankPath != null) {
        log.info("Reading in trees from " + testTreebankPath);
        if (testTreebankFilter != null) {
            log.info("Filtering on " + testTreebankFilter);
        }
        testTreebank = dvparser.getOp().tlpParams.memoryTreebank();
        ;
        testTreebank.loadPath(testTreebankPath, testTreebankFilter);
        log.info("Read in " + testTreebank.size() + " trees for testing");
    }
    //    runGradientCheck= true;
    if (runGradientCheck) {
        log.info("Running gradient check on " + trainSentences.size() + " trees");
        dvparser.runGradientCheck(trainSentences, trainCompressedParses);
    }
    if (runTraining) {
        log.info("Training the RNN parser");
        log.info("Current train options: " + dvparser.getOp().trainOptions);
        dvparser.train(trainSentences, trainCompressedParses, testTreebank, modelPath, resultsRecordPath);
        if (modelPath != null) {
            dvparser.saveModel(modelPath);
        }
    }
    if (testTreebankPath != null) {
        EvaluateTreebank evaluator = new EvaluateTreebank(dvparser.attachModelToLexicalizedParser());
        evaluator.testOnTreebank(testTreebank);
    }
    log.info("Successfully ran DVParser");
}

Example 10 with TreeTransformer

use of edu.stanford.nlp.trees.TreeTransformer in project CoreNLP by stanfordnlp.

the class FTBCorrector method main.

/**
   * @param args
   */
public static void main(String[] args) {
    if (args.length != 1) {
        log.info("Usage: java " + FTBCorrector.class.getName() + " filename\n");
        System.exit(-1);
    }
    TreeTransformer tt = new FTBCorrector();
    File f = new File(args[0]);
    try {
        //These bad trees in the Candito training set should be thrown out:
        //  (ROOT (SENT (" ") (. .)))
        //  (ROOT (SENT (. .)))
        TregexPattern pBadTree = TregexPattern.compile("@SENT <: @PUNC");
        TregexPattern pBadTree2 = TregexPattern.compile("@SENT <1 @PUNC <2 @PUNC !<3 __");
        BufferedReader br = new BufferedReader(new InputStreamReader(new FileInputStream(f), "UTF-8"));
        TreeReaderFactory trf = new FrenchTreeReaderFactory();
        TreeReader tr = trf.newTreeReader(br);
        int nTrees = 0;
        for (Tree t; (t = tr.readTree()) != null; nTrees++) {
            TregexMatcher m = pBadTree.matcher(t);
            TregexMatcher m2 = pBadTree2.matcher(t);
            if (m.find() || m2.find()) {
                log.info("Discarding tree: " + t.toString());
            } else {
                Tree fixedT = tt.transformTree(t);
                System.out.println(fixedT.toString());
            }
        }
        tr.close();
        System.err.printf("Wrote %d trees%n", nTrees);
    } catch (UnsupportedEncodingException e) {
        e.printStackTrace();
    } catch (FileNotFoundException e) {
        e.printStackTrace();
    } catch (IOException e) {
        e.printStackTrace();
    } catch (TregexParseException e) {
        e.printStackTrace();
    }
}