Exemple #1
0
  /** @param args */
  public static void main(String[] args) {
    if (args.length != 1) {
      System.err.println("Usage: java " + ATBCorrector.class.getName() + " filename\n");
      System.exit(-1);
    }

    TreeTransformer tt = new ATBCorrector();

    File f = new File(args[0]);
    try {

      BufferedReader br =
          new BufferedReader(new InputStreamReader(new FileInputStream(f), "UTF-8"));
      TreeReaderFactory trf = new ArabicTreeReaderFactory.ArabicRawTreeReaderFactory();
      TreeReader tr = trf.newTreeReader(br);

      int nTrees = 0;
      for (Tree t; (t = tr.readTree()) != null; nTrees++) {
        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 (IOException e) {
      e.printStackTrace();
    }
  }
 /**
  * Returns the set of dependencies in a tree, according to some {@link
  * edu.stanford.nlp.trees.DependencyTyper}.
  */
 public static <E> Collection<E> dependencyObjectify(
     Tree t, HeadFinder hf, TreeTransformer collinizer, DependencyTyper<E> typer) {
   Collection<E> deps = new ArrayList<E>();
   Tree t1 = collinizer.transformTree(t);
   if (t1 == null) return deps;
   dependencyObjectifyHelper(t1, t1, hf, deps, typer);
   return deps;
 }
 List<Tree> prune(List<Tree> treeList, Label label, int start, int end) {
   // get reference tree
   if (treeList.size() == 1) {
     return treeList;
   }
   Tree testTree = treeList.get(0).treeFactory().newTreeNode(label, treeList);
   int goal = Numberer.getGlobalNumberer("states").number(label.value());
   Tree tempTree = parser.extractBestParse(goal, start, end);
   // parser.restoreUnaries(tempTree);
   Tree pcfgTree = debinarizer.transformTree(tempTree);
   Set<Constituent> pcfgConstituents =
       pcfgTree.constituents(new LabeledScoredConstituentFactory());
   // delete child labels that are not in reference but do not cross reference
   List<Tree> prunedChildren = new ArrayList<Tree>();
   int childStart = 0;
   for (int c = 0, numCh = testTree.numChildren(); c < numCh; c++) {
     Tree child = testTree.getChild(c);
     boolean isExtra = true;
     int childEnd = childStart + child.yield().size();
     Constituent childConstituent =
         new LabeledScoredConstituent(childStart, childEnd, child.label(), 0);
     if (pcfgConstituents.contains(childConstituent)) {
       isExtra = false;
     }
     if (childConstituent.crosses(pcfgConstituents)) {
       isExtra = false;
     }
     if (child.isLeaf() || child.isPreTerminal()) {
       isExtra = false;
     }
     if (pcfgTree.yield().size() != testTree.yield().size()) {
       isExtra = false;
     }
     if (!label.value().startsWith("NP^NP")) {
       isExtra = false;
     }
     if (isExtra) {
       System.err.println(
           "Pruning: "
               + child.label()
               + " from "
               + (childStart + start)
               + " to "
               + (childEnd + start));
       System.err.println("Was: " + testTree + " vs " + pcfgTree);
       prunedChildren.addAll(child.getChildrenAsList());
     } else {
       prunedChildren.add(child);
     }
     childStart = childEnd;
   }
   return prunedChildren;
 }
 /**
  * Takes a Tree and a collinizer and returns a Collection of {@link Constituent}s for PARSEVAL
  * evaluation. Some notes on this particular parseval:
  *
  * <ul>
  *   <li>It is character-based, which allows it to be used on segmentation/parsing combination
  *       evaluation.
  *   <li>whether it gives you labeled or unlabeled bracketings depends on the value of the <code>
  *       labelConstituents</code> parameter
  * </ul>
  *
  * (Note that I haven't checked this rigorously yet with the PARSEVAL definition -- Roger.)
  */
 public static Collection<Constituent> parsevalObjectify(
     Tree t, TreeTransformer collinizer, boolean labelConstituents) {
   Collection<Constituent> spans = new ArrayList<Constituent>();
   Tree t1 = collinizer.transformTree(t);
   if (t1 == null) {
     return spans;
   }
   for (Tree node : t1) {
     if (node.isLeaf() || node.isPreTerminal() || (node != t1 && node.parent(t1) == null)) {
       continue;
     }
     int leftEdge = t1.leftCharEdge(node);
     int rightEdge = t1.rightCharEdge(node);
     if (labelConstituents) spans.add(new LabeledConstituent(leftEdge, rightEdge, node.label()));
     else spans.add(new SimpleConstituent(leftEdge, rightEdge));
   }
   return spans;
 }
  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.

    System.out.println(StringUtils.toInvocationString("FactoredParser", 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) {
          System.err.println("Class not found: " + args[i + 1]);
          throw new RuntimeException(e);
        } catch (InstantiationException e) {
          System.err.println("Couldn't instantiate: " + args[i + 1] + ": " + e.toString());
          throw new RuntimeException(e);
        } catch (IllegalAccessException e) {
          System.err.println("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 =
        new HashSet<String>(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();
    System.err.print("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.");

    System.err.print("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<Tree>();

    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<String>();
        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);
            }
          }
        }
        System.err.println("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<Tree>();
    for (Tree tree : testTreebank) {
      if (op.trainOptions.collinsPunc) {
        tree = collinsPuncTransformer.transformTree(tree);
      }
      tree = binarizer.transformTree(tree);
      binaryTestTrees.add(tree);
    }
    Timing.tick("done."); // binarization
    BinaryGrammar bg = null;
    UnaryGrammar ug = null;
    DependencyGrammar dg = null;
    // DependencyGrammar dgBLIPP = null;
    Lexicon lex = null;
    Index<String> stateIndex = new HashIndex<String>();

    // extract grammars
    Extractor<Pair<UnaryGrammar, BinaryGrammar>> bgExtractor =
        new BinaryGrammarExtractor(op, stateIndex);
    // Extractor bgExtractor = new SmoothedBinaryGrammarExtractor();//new BinaryGrammarExtractor();
    // Extractor lexExtractor = new LexiconExtractor();

    // Extractor dgExtractor = new DependencyMemGrammarExtractor();

    if (op.doPCFG) {
      System.err.print("Extracting PCFG...");
      Pair<UnaryGrammar, BinaryGrammar> bgug = null;
      if (op.trainOptions.cheatPCFG) {
        List<Tree> allTrees = new ArrayList<Tree>(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.");
    }
    System.err.print("Extracting Lexicon...");
    Index<String> wordIndex = new HashIndex<String>();
    Index<String> tagIndex = new HashIndex<String>();
    lex = op.tlpParams.lex(op, wordIndex, tagIndex);
    lex.train(binaryTrainTrees);
    Timing.tick("done.");

    if (op.doDep) {
      System.err.print("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);
      dg =
          dgExtractor.extract(
              binaryTrainTrees); // uses information whether the words are known or not, discards
      // unknown words
      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) {
      System.err.print("Serializing parser...");
      LexicalizedParser.saveParserDataToSerialized(
          new ParserData(lex, bg, ug, dg, stateIndex, wordIndex, tagIndex, op), 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) {
        System.err.println(e);
        System.err.println("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) {
        // System.err.println(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;
      System.err.println("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) {
          System.err.println("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);
      } // end if doPCFG

      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));
        }
      }
    } // end for each tree in test treebank

    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();
  }
  /**
   * Run the Evalb scoring metric on guess/gold input. The default language is English.
   *
   * @param args
   */
  public static void main(String[] args) {
    TreebankLangParserParams tlpp = new EnglishTreebankParserParams();
    int maxGoldYield = Integer.MAX_VALUE;
    boolean VERBOSE = false;
    String encoding = "UTF-8";

    String guessFile = null;
    String goldFile = null;

    Map<String, String[]> argsMap = StringUtils.argsToMap(args, optionArgDefs);

    for (Map.Entry<String, String[]> opt : argsMap.entrySet()) {
      if (opt.getKey() == null) continue;
      if (opt.getKey().equals("-l")) {
        Language lang = Language.valueOf(opt.getValue()[0].trim());
        tlpp = lang.params;

      } else if (opt.getKey().equals("-y")) {
        maxGoldYield = Integer.parseInt(opt.getValue()[0].trim());

      } else if (opt.getKey().equals("-v")) {
        VERBOSE = true;

      } else if (opt.getKey().equals("-e")) {
        encoding = opt.getValue()[0];

      } else {
        System.err.println(usage.toString());
        System.exit(-1);
      }

      // Non-option arguments located at key null
      String[] rest = argsMap.get(null);
      if (rest == null || rest.length < minArgs) {
        System.err.println(usage.toString());
        System.exit(-1);
      }
      goldFile = rest[0];
      guessFile = rest[1];
    }

    tlpp.setInputEncoding(encoding);
    final PrintWriter pwOut = tlpp.pw();

    final Treebank guessTreebank = tlpp.diskTreebank();
    guessTreebank.loadPath(guessFile);
    pwOut.println("GUESS TREEBANK:");
    pwOut.println(guessTreebank.textualSummary());

    final Treebank goldTreebank = tlpp.diskTreebank();
    goldTreebank.loadPath(goldFile);
    pwOut.println("GOLD TREEBANK:");
    pwOut.println(goldTreebank.textualSummary());

    final UnlabeledAttachmentEval metric =
        new UnlabeledAttachmentEval("UAS LP/LR", true, tlpp.headFinder());

    final TreeTransformer tc = tlpp.collinizer();

    // The evalb ref implementation assigns status for each tree pair as follows:
    //
    //   0 - Ok (yields match)
    //   1 - length mismatch
    //   2 - null parse e.g. (()).
    //
    // In the cases of 1,2, evalb does not include the tree pair in the LP/LR computation.
    final Iterator<Tree> goldItr = goldTreebank.iterator();
    final Iterator<Tree> guessItr = guessTreebank.iterator();
    int goldLineId = 0;
    int guessLineId = 0;
    int skippedGuessTrees = 0;
    while (guessItr.hasNext() && goldItr.hasNext()) {
      Tree guessTree = guessItr.next();
      List<? extends Label> guessYield = guessTree.yield();
      guessLineId++;

      Tree goldTree = goldItr.next();
      List<? extends Label> goldYield = goldTree.yield();
      goldLineId++;

      // Check that we should evaluate this tree
      if (goldYield.size() > maxGoldYield) {
        skippedGuessTrees++;
        continue;
      }

      // Only trees with equal yields can be evaluated
      if (goldYield.size() != guessYield.size()) {
        pwOut.printf(
            "Yield mismatch gold: %d tokens vs. guess: %d tokens (lines: gold %d guess %d)%n",
            goldYield.size(), guessYield.size(), goldLineId, guessLineId);
        skippedGuessTrees++;
        continue;
      }

      final Tree evalGuess = tc.transformTree(guessTree);
      evalGuess.indexLeaves(true);
      final Tree evalGold = tc.transformTree(goldTree);
      evalGold.indexLeaves(true);

      metric.evaluate(evalGuess, evalGold, ((VERBOSE) ? pwOut : null));
    }

    if (guessItr.hasNext() || goldItr.hasNext()) {
      System.err.printf(
          "Guess/gold files do not have equal lengths (guess: %d gold: %d)%n.",
          guessLineId, goldLineId);
    }

    pwOut.println(
        "================================================================================");
    if (skippedGuessTrees != 0)
      pwOut.printf("%s %d guess trees\n", "Unable to evaluate", skippedGuessTrees);
    metric.display(true, pwOut);

    pwOut.println();
    pwOut.close();
  }
  /** Execute with no arguments for usage. */
  public static void main(String[] args) {

    if (!validateCommandLine(args)) {
      System.err.println(USAGE);
      System.exit(-1);
    }

    final TreebankLangParserParams tlpp = LANGUAGE.params;
    final PrintWriter pwOut = tlpp.pw();

    final Treebank guessTreebank = tlpp.diskTreebank();
    guessTreebank.loadPath(guessFile);
    pwOut.println("GUESS TREEBANK:");
    pwOut.println(guessTreebank.textualSummary());

    final Treebank goldTreebank = tlpp.diskTreebank();
    goldTreebank.loadPath(goldFile);
    pwOut.println("GOLD TREEBANK:");
    pwOut.println(goldTreebank.textualSummary());

    final LeafAncestorEval metric = new LeafAncestorEval("LeafAncestor");

    final TreeTransformer tc = tlpp.collinizer();

    // The evalb ref implementation assigns status for each tree pair as follows:
    //
    //   0 - Ok (yields match)
    //   1 - length mismatch
    //   2 - null parse e.g. (()).
    //
    // In the cases of 1,2, evalb does not include the tree pair in the LP/LR computation.
    final Iterator<Tree> goldItr = goldTreebank.iterator();
    final Iterator<Tree> guessItr = guessTreebank.iterator();
    int goldLineId = 0;
    int guessLineId = 0;
    int skippedGuessTrees = 0;
    while (guessItr.hasNext() && goldItr.hasNext()) {
      Tree guessTree = guessItr.next();
      List<? extends Label> guessYield = guessTree.yield();
      guessLineId++;

      Tree goldTree = goldItr.next();
      List<? extends Label> goldYield = goldTree.yield();
      goldLineId++;

      // Check that we should evaluate this tree
      if (goldYield.size() > MAX_GOLD_YIELD) {
        skippedGuessTrees++;
        continue;
      }

      // Only trees with equal yields can be evaluated
      if (goldYield.size() != guessYield.size()) {
        pwOut.printf(
            "Yield mismatch gold: %d tokens vs. guess: %d tokens (lines: gold %d guess %d)%n",
            goldYield.size(), guessYield.size(), goldLineId, guessLineId);
        skippedGuessTrees++;
        continue;
      }

      final Tree evalGuess = tc.transformTree(guessTree);
      final Tree evalGold = tc.transformTree(goldTree);

      metric.evaluate(evalGuess, evalGold, ((VERBOSE) ? pwOut : null));
    }

    if (guessItr.hasNext() || goldItr.hasNext()) {
      System.err.printf(
          "Guess/gold files do not have equal lengths (guess: %d gold: %d)%n.",
          guessLineId, goldLineId);
    }

    pwOut.println(
        "================================================================================");
    if (skippedGuessTrees != 0)
      pwOut.printf("%s %d guess trees%n", "Unable to evaluate", skippedGuessTrees);
    metric.display(true, pwOut);
    pwOut.close();
  }