Stanford parser学习：LexicalizedParser类分析

上次（http://www.cnblogs.com/stGeekpower/p/3457746.html）主要是对应于javadoc写了下LexicalizedParser类main函数的功能，这次看下main函数的具体处理过程。main函数大概350行左右，主要完成的工作是：初始化变量（各种标志位）、解析传入的各种参数、根据传入的选项参数分步骤完成各种工作。

根据选项来做的工作按顺序主要包括：分词（必须最先处理）、初始化LexicalizedParser（读入或训练）、编码设置、测试、保存（如果需要的话）、解析输出结果。

具体解析的部分：对句子解析是通过LexicalizedParser对象生成的ParserQuery类的parse函数来完成，对文件的解析由ParseFiles类的parseFiles函数（最终也是调用ParserQuery类）完成。

一、初始化变量

这部分主要处理申明一些标志位，以及构建解析器需要的变量；

 boolean train = false;//train or parse
 boolean saveToSerializedFile = false;//是否序列化存储至文件
 boolean saveToTextFile = false;//是否存储至文本文件
 String serializedInputFileOrUrl = null;//序列化输入文件或者url
 String textInputFileOrUrl = null;//文本输入文件或者url
 String serializedOutputFileOrUrl = null;//序列化输出文件或者url
 String textOutputFileOrUrl = null;//文本输入文件或者url
 String treebankPath = null;//语法树路径
 Treebank testTreebank = null;
 Treebank tuneTreebank = null;
 String testPath = null;
 FileFilter testFilter = null;
 String tunePath = null;
 FileFilter tuneFilter = null;
 FileFilter trainFilter = null;//训练过滤范围
 String secondaryTreebankPath = null;
double secondaryTreebankWeight = 1.0;
 FileFilter secondaryTrainFilter = null;
 
 // variables needed to process the files to be parsed
 TokenizerFactory<? extends HasWord> tokenizerFactory = null; //分词工厂
 String tokenizerOptions = null;//分词所需参数
 String tokenizerFactoryClass = null;//分词所用类
 String tokenizerMethod = null;//分词所用方法
 boolean tokenized = false; // whether or not the input file has already been tokenized
 Function<List<HasWord>, List<HasWord>> escaper = null; //转义
 String tagDelimiter = null; //分隔符
 String sentenceDelimiter = null;
 String elementDelimiter = null;

二、解析传入的各种参数

这里处理用户传入的各种选项参数，存入在一种申明的变量中；

 int argIndex = 0;
 if (args.length < 1) {//参数数量为0，错误返回
      System.err.println("Basic usage (see Javadoc for more): java edu.stanford.nlp.parser.lexparser" +
                    ".LexicalizedParser parserFileOrUrl filename*");
     return;
 }
 
 Options op = new Options(); //处理参数的对象
 List<String> optionArgs = new ArrayList<String>();
 String encoding = null;
 // while loop through option arguments,循环处理选项参数
 while (argIndex < args.length && args[argIndex].charAt(0) == '-') {
     if (args[argIndex].equalsIgnoreCase("-train") || args[argIndex].equalsIgnoreCase("-trainTreebank")) {//判断是否执行训练功能
          train = true;
         //处理训练时传入的参数信息，得到文件路径和过滤范围存至treebankDescription
         Pair<String, FileFilter> treebankDescription = ArgUtils.getTreebankDescription(args, argIndex, "-test");
         argIndex = argIndex + ArgUtils.numSubArgs(args, argIndex) + 1;
         treebankPath = treebankDescription.first();
         trainFilter = treebankDescription.second();
         } else if (args[argIndex].equalsIgnoreCase("-train2")) {
             // TODO: we could use the fully expressive -train options if
             // we add some mechanism for returning leftover options from
             // ArgUtils.getTreebankDescription
             // train = true;     // cdm july 2005: should require -train for this
             int numSubArgs = ArgUtils.numSubArgs(args, argIndex);
             argIndex++;
             if (numSubArgs < 2) {
                 throw new RuntimeException("Error: -train2 <treebankPath> [<ranges>] <weight>.");
             }
             secondaryTreebankPath = args[argIndex++];
             secondaryTrainFilter = (numSubArgs == 3) ? new NumberRangesFileFilter(args[argIndex++], true) : null;
             secondaryTreebankWeight = Double.parseDouble(args[argIndex++]);
         } else if (args[argIndex].equalsIgnoreCase("-tLPP") && (argIndex + 1 < args.length)) {
              // 当使用除英文外的语言或者English Penn Treebank之外的Treebank时候需要指定TreebankLangParserParams，
                // 该选项必须出现在其他的与语言相关的选项之前。不同的语言有不同的参数
                try {
                    op.tlpParams = (TreebankLangParserParams) Class.forName(args[argIndex + 1]).newInstance();
              } catch (ClassNotFoundException e) {
                  System.err.println("Class not found: " + args[argIndex + 1]);
                  throw new RuntimeException(e);
              } catch (InstantiationException e) {
                  System.err.println("Couldn't instantiate: " + args[argIndex + 1] + ": " + e.toString());
                  throw new RuntimeException(e);
              } catch (IllegalAccessException e) {
                  System.err.println("Illegal access" + e);
                  throw new RuntimeException(e);
              }
              argIndex += 2;
          } else if (args[argIndex].equalsIgnoreCase("-encoding")) {//编码
                // sets encoding for TreebankLangParserParams
              // redone later to override any serialized parser one read in
              encoding = args[argIndex + 1];
              op.tlpParams.setInputEncoding(encoding);
              op.tlpParams.setOutputEncoding(encoding);
              argIndex += 2;
          } else if (args[argIndex].equalsIgnoreCase("-tokenized")) {//是否已经分词
                 tokenized = true;
               argIndex += 1;
          } else if (args[argIndex].equalsIgnoreCase("-escaper")) {
               try {
                   escaper = ReflectionLoading.loadByReflection(args[argIndex + 1]);
               } catch (Exception e) {
                   System.err.println("Couldn't instantiate escaper " + args[argIndex + 1] + ": " + e);
               }
               argIndex += 2;
          } else if (args[argIndex].equalsIgnoreCase("-tokenizerOptions")) {//指定TokenizerFactory类完成tokenization 所需要的参数信息
                tokenizerOptions = args[argIndex + 1];
              argIndex += 2;
          } else if (args[argIndex].equalsIgnoreCase("-tokenizerFactory")) {//指定一个TokenizerFactory类来完成分词
                tokenizerFactoryClass = args[argIndex + 1];
              argIndex += 2;
          } else if (args[argIndex].equalsIgnoreCase("-tokenizerMethod")) {//分词方法
                tokenizerMethod = args[argIndex + 1];
              argIndex += 2;
          } else if (args[argIndex].equalsIgnoreCase("-sentences")) {//指定一个词语来划分句子边界,即分句根据
                sentenceDelimiter = args[argIndex + 1];
              if (sentenceDelimiter.equalsIgnoreCase("newline")) {
                  sentenceDelimiter = "\n";
              }
              argIndex += 2;
          } else if (args[argIndex].equalsIgnoreCase("-parseInside")) {//解析的范围，可以是句，几句等等
                elementDelimiter = args[argIndex + 1];
              argIndex += 2;
          } else if (args[argIndex].equalsIgnoreCase("-tagSeparator")) {//指明标注符号
                tagDelimiter = args[argIndex + 1];
              argIndex += 2;
          } else if (args[argIndex].equalsIgnoreCase("-loadFromSerializedFile") ||
              args[argIndex].equalsIgnoreCase("-model")) {
              // load the parser from a binary serialized file
              // the next argument must be the path to the parser file
              serializedInputFileOrUrl = args[argIndex + 1];
              argIndex += 2;
          } else if (args[argIndex].equalsIgnoreCase("-loadFromTextFile")) {
              // load the parser from declarative text file
              // the next argument must be the path to the parser file
              textInputFileOrUrl = args[argIndex + 1];
              argIndex += 2;
          } else if (args[argIndex].equalsIgnoreCase("-saveToSerializedFile")) {
              saveToSerializedFile = true;
              if (ArgUtils.numSubArgs(args, argIndex) < 1) {
                  System.err.println("Missing path: -saveToSerialized filename");
              } else {
                  serializedOutputFileOrUrl = args[argIndex + 1];
              }
              argIndex += 2;
          } else if (args[argIndex].equalsIgnoreCase("-saveToTextFile")) {
              // save the parser to declarative text file
              saveToTextFile = true;
              textOutputFileOrUrl = args[argIndex + 1];
              argIndex += 2;
          } else if (args[argIndex].equalsIgnoreCase("-saveTrainTrees")) {
              // save the training trees to a binary file
              op.trainOptions.trainTreeFile = args[argIndex + 1];
              argIndex += 2;
          } else if (args[argIndex].equalsIgnoreCase("-treebank") ||
              args[argIndex].equalsIgnoreCase("-testTreebank") ||
              args[argIndex].equalsIgnoreCase("-test")) {//训练并测试，测试所需的参数
                Pair<String, FileFilter> treebankDescription = ArgUtils.getTreebankDescription(args, argIndex, "-test");
              argIndex = argIndex + ArgUtils.numSubArgs(args, argIndex) + 1;
              testPath = treebankDescription.first();
              testFilter = treebankDescription.second();
          } else if (args[argIndex].equalsIgnoreCase("-tune")) {
              Pair<String, FileFilter> treebankDescription = ArgUtils.getTreebankDescription(args, argIndex, "-tune");
              argIndex = argIndex + ArgUtils.numSubArgs(args, argIndex) + 1;
              tunePath = treebankDescription.first();
              tuneFilter = treebankDescription.second();
          } else {
              int oldIndex = argIndex;
              argIndex = op.setOptionOrWarn(args, argIndex);
              for (int i = oldIndex; i < argIndex; i++) {
                  optionArgs.add(args[i]);
              }
          }
 } // end while loop through arguments

三、分词处理

句法分析的前提是句子已经被正确分词，这里即完成分词工作，当然分词我们可以选用自己合适的分词器；

// set up tokenizerFactory with options if provided
        if (tokenizerFactoryClass != null || tokenizerOptions != null) {
            try {//分词工厂类、分词方法由参数指定，若不指定，默认PTBTokenizer
                if (tokenizerFactoryClass != null) {
                    Class<TokenizerFactory<? extends HasWord>> clazz = ErasureUtils.uncheckedCast(Class.forName
                            (tokenizerFactoryClass));
                    Method factoryMethod;
                    if (tokenizerOptions != null) {
                        factoryMethod = clazz.getMethod(tokenizerMethod != null ? tokenizerMethod :
                                "newWordTokenizerFactory", String.class);
                        tokenizerFactory = ErasureUtils.uncheckedCast(factoryMethod.invoke(null, tokenizerOptions));
                    } else {
                        factoryMethod = clazz.getMethod(tokenizerMethod != null ? tokenizerMethod :
                                "newTokenizerFactory");
                        tokenizerFactory = ErasureUtils.uncheckedCast(factoryMethod.invoke(null));
                    }
                } else {
                    // have options but no tokenizer factory; default to PTB
                    tokenizerFactory = PTBTokenizer.PTBTokenizerFactory.newWordTokenizerFactory(tokenizerOptions);
                }
            } catch (IllegalAccessException e) {
                System.err.println("Couldn't instantiate TokenizerFactory " + tokenizerFactoryClass + " with options " +
                        "" + tokenizerOptions);
                throw new RuntimeException(e);
            } catch (NoSuchMethodException e) {
                System.err.println("Couldn't instantiate TokenizerFactory " + tokenizerFactoryClass + " with options " +
                        "" + tokenizerOptions);
                throw new RuntimeException(e);
            } catch (ClassNotFoundException e) {
                System.err.println("Couldn't instantiate TokenizerFactory " + tokenizerFactoryClass + " with options " +
                        "" + tokenizerOptions);
                throw new RuntimeException(e);
            } catch (InvocationTargetException e) {
                System.err.println("Couldn't instantiate TokenizerFactory " + tokenizerFactoryClass + " with options " +
                        "" + tokenizerOptions);
                throw new RuntimeException(e);
            }

四、初始化LexicalizedParser

初始化LexicalizedParser有三种方式，分别是：根据数据训练一个，从文本文件读入，从序列化文件读入；

if (tuneFilter != null || tunePath != null) {//处理tune treebank
            if (tunePath == null) {
                if (treebankPath == null) {
                    throw new RuntimeException("No tune treebank path specified...");
                } else {
                    System.err.println("No tune treebank path specified.  Using train path: \"" + treebankPath + '\"');
                    tunePath = treebankPath;
                }
            }
            tuneTreebank = op.tlpParams.testMemoryTreebank();
            tuneTreebank.loadPath(tunePath, tuneFilter);
        }
 
        if (!train && op.testOptions.verbose) {
            StringUtils.printErrInvocationString("LexicalizedParser", args);
        }
        edu.stanford.nlp.parser.lexparser.LexicalizedParser lp; // always initialized in next if-then-else block
        if (train) {
            StringUtils.printErrInvocationString("LexicalizedParser", args);
 
            // so we train a parser using the treebank
            GrammarCompactor compactor = null;
            if (op.trainOptions.compactGrammar() == 3) {
                compactor = new ExactGrammarCompactor(op, false, false);
            }
 
            Treebank trainTreebank = makeTreebank(treebankPath, op, trainFilter);
 
            Treebank secondaryTrainTreebank = null;
            if (secondaryTreebankPath != null) {
                secondaryTrainTreebank = makeSecondaryTreebank(secondaryTreebankPath, op, secondaryTrainFilter);
            }
 
            List<List<TaggedWord>> extraTaggedWords = null;
            if (op.trainOptions.taggedFiles != null) {
                extraTaggedWords = new ArrayList<List<TaggedWord>>();
                List<TaggedFileRecord> fileRecords = TaggedFileRecord.createRecords(new Properties(),
                        op.trainOptions.taggedFiles);
                for (TaggedFileRecord record : fileRecords) {
                    for (List<TaggedWord> sentence : record.reader()) {
                        extraTaggedWords.add(sentence);
                    }
                }
            }
            //执行训练方法时对lp的初始化，根据标注数据训练出lp
            lp = getParserFromTreebank(trainTreebank, secondaryTrainTreebank, secondaryTreebankWeight, compactor, op,
                    tuneTreebank, extraTaggedWords);
        } else if (textInputFileOrUrl != null) {
            // so we load the parser from a text grammar file,直接从文本文件中读入lp
            lp = getParserFromTextFile(textInputFileOrUrl, op);
        } else {
            // so we load a serialized parser,从序列化保存的文件中读入lp
            if (serializedInputFileOrUrl == null && argIndex < args.length) {
                // the next argument must be the path to the serialized parser
                serializedInputFileOrUrl = args[argIndex];
                argIndex++;
            }
            if (serializedInputFileOrUrl == null) {
                System.err.println("No grammar specified, exiting...");
                return;
            }
            String[] extraArgs = new String[optionArgs.size()];
            extraArgs = optionArgs.toArray(extraArgs);
            try {
                lp = loadModel(serializedInputFileOrUrl, op, extraArgs);
                op = lp.op;
            } catch (IllegalArgumentException e) {
                System.err.println("Error loading parser, exiting...");
                throw e;
            }
        }

五、控制编码

 // the following has to go after reading parser to make sure
 // op and tlpParams are the same for train and test
 // THIS IS BUTT UGLY BUT IT STOPS USER SPECIFIED ENCODING BEING
 // OVERWRITTEN BY ONE SPECIFIED IN SERIALIZED PARSER
 if (encoding != null) {
    op.tlpParams.setInputEncoding(encoding);
    op.tlpParams.setOutputEncoding(encoding);
 }

六、测试数据设置

       if (testFilter != null || testPath != null) {
            if (testPath == null) {
                if (treebankPath == null) {
                    throw new RuntimeException("No test treebank path specified...");
                } else {
                    System.err.println("No test treebank path specified.  Using train path: \"" + treebankPath + '\"');
                    testPath = treebankPath;
                }
            }
            testTreebank = op.tlpParams.testMemoryTreebank();
            testTreebank.loadPath(testPath, testFilter);
        }

七、需要的话将训练生成的解析器保存

        op.trainOptions.sisterSplitters = Generics.newHashSet(Arrays.asList(op.tlpParams.sisterSplitters()));
 
        // at this point we should be sure that op.tlpParams is
        // set appropriately (from command line, or from grammar file),
        // and will never change again.  -- Roger
 
        // Now what do we do with the parser we've made
        if (saveToTextFile) {
            // save the parser to textGrammar format
            if (textOutputFileOrUrl != null) {
                lp.saveParserToTextFile(textOutputFileOrUrl);
            } else {
                System.err.println("Usage: must specify a text grammar output path");
            }
        }
        if (saveToSerializedFile) {
            if (serializedOutputFileOrUrl != null) {
                lp.saveParserToSerialized(serializedOutputFileOrUrl);
            } else if (textOutputFileOrUrl == null && testTreebank == null) {
                // no saving/parsing request has been specified
                System.err.println("usage: " + "java edu.stanford.nlp.parser.lexparser.LexicalizedParser " + "-train " +
                        "trainFilesPath [fileRange] -saveToSerializedFile serializedParserFilename");
            }
        }

八、训练或者指定输入参数时，输出一些信息

        if (op.testOptions.verbose || train) {
            // Tell the user a little or a lot about what we have made
            // get lexicon size separately as it may have its own prints in it....
            String lexNumRules = lp.lex != null ? Integer.toString(lp.lex.numRules()) : "";
            System.err.println("Grammar\tStates\tTags\tWords\tUnaryR\tBinaryR\tTaggings");
            System.err.println("Grammar\t" +
                    lp.stateIndex.size() + '\t' +
                    lp.tagIndex.size() + '\t' +
                    lp.wordIndex.size() + '\t' +
                    (lp.ug != null ? lp.ug.numRules() : "") + '\t' +
                    (lp.bg != null ? lp.bg.numRules() : "") + '\t' +
                    lexNumRules);
            System.err.println("ParserPack is " + op.tlpParams.getClass().getName());
            System.err.println("Lexicon is " + lp.lex.getClass().getName());
            if (op.testOptions.verbose) {
                System.err.println("Tags are: " + lp.tagIndex);
                // System.err.println("States are: " + lp.pd.stateIndex); // This is too verbose. It was already
                // printed out by the below printOptions command if the flag -printStates is given (at training time)!
            }
            printOptions(false, op);
        }

九、执行解析工作

可以以句子的方式解析，也可用ParseFiles类的方法来解析多个文件。

        if (testTreebank != null) {
            // test parser on treebank
            EvaluateTreebank evaluator = new EvaluateTreebank(lp);
            evaluator.testOnTreebank(testTreebank);
        } else if (argIndex >= args.length) {
            // no more arguments, so we just parse our own test sentence
            PrintWriter pwOut = op.tlpParams.pw();
            PrintWriter pwErr = op.tlpParams.pw(System.err);
            ParserQuery pq = lp.parserQuery();
            if (pq.parse(op.tlpParams.defaultTestSentence())) {//解析
                lp.getTreePrint().printTree(pq.getBestParse(), pwOut);
            } else {
                pwErr.println("Error. Can't parse test sentence: " +
                        op.tlpParams.defaultTestSentence());
            }
        } else {
            // We parse filenames given by the remaining arguments，解析
            ParseFiles.parseFiles(args, argIndex, tokenized, tokenizerFactory, elementDelimiter, sentenceDelimiter,
                    escaper, tagDelimiter, op, lp.getTreePrint(), lp);
        }