我分词想在你们模型的基础上增加一些自定义词库。。这个怎么处理呢？

以下列表中的文件LTP已经不再使用

__util/conversion_utf.h
__util/decode_gbk.h
__util/EncodeUtil.cpp
__util/EncodeUtil.h
__util/gbk_u16.h
__util/IniReader.cpp
__util/IniReader.h
__util/Logger.cpp
__util/Logger.h
__util/md5.cpp
__util/md5.h
__util/SBC2DBC.cpp
__util/SBC2DBC.h
__util/TextProcess.cpp
__util/TextProcess.h
__util/Timer.h

extractord单件在运行结束时没有合理释放，valgrind memcheck报告内存泄露。
单件正确的实现方法应参考：http://stackoverflow.com/questions/1008019/c-singleton-design-pattern

current performance of ltp-server is list below.

• average response time : 1.25 (secs)
• throughput : 7.806 (req/sec)

here goes some problems:

• cpu usage is low during the test (about 60% single thread). maybe too much time is consumed by web-io?
• why not implement a multithreaded server to improve performance.

我的代码如下:
作者能提供一个srl调用例子么？

/*
* NodeJS LTP 扩展
* @author 蜗眼 
* 第一次写C/C++ 见笑了。。。
*/
#include < v8.h > #include "Xml4nlp.h"#include "Ltp.h"#include "segment_dll.h"#include "postag_dll.h"#include "ner_dll.h"#include "parser_dll.h"#include "SRL_DLL.h"#include < iostream > #include < string > #include < node.h > using namespace node; 
using namespace v8; 
using namespace std; 
//using namespace ltp::strutils::codecs;
#include "DepSRL.h"static DepSRL g_depSRL; 
Handle < Value > Method(const Arguments & args) {
   HandleScope scope; 
   const char * path = "/Users/iceet/Mine/Bill/tools/xls/ltp_data/cws.model"; 
   const char * selfs = "/Users/iceet/Mine/Bill/src/BillNLP/self.dic"; 
   void * engine = segmentor_create_segmentor(path, 
   selfs); 
   vector < string > words; 
   //分词接口
   int len = segmentor_segment(engine, 
   "我们都是**人。", words); 
   for (int i = 0; i < len; ++ i) {
      std :: cout << words[i] << "|"; 
      }
   std :: cout << std :: endl; 
   segmentor_release_segmentor(engine); 
   //词性标注
   void * engine1 = postagger_create_postagger("/Users/iceet/Mine/Bill/tools/xls/ltp_data/pos.model"); 
   std :: vector < std :: string > tags; 
   postagger_postag(engine1, words, tags); 
   for (int i = 0; i < tags.size(); ++ i) {
      std :: cout << words[i] << "/" << tags[i]; 
      if (i == tags.size() - 1) std :: cout << std :: endl; 
      else std :: cout << " "; 
      }
   postagger_release_postagger(engine1); 
   //命名实体识别接口
   void * engin2 = ner_create_recognizer("/Users/iceet/Mine/Bill/tools/xls/ltp_data/ner.model"); 
   int ret; 
   std :: vector < string > vec; 
   ret = ner_recognize(engin2, words, tags, vec); 
   //std::cont << vec.size() <<std::endl;
   for (int i = 0; i < vec.size(); ++ i) {
      std :: cout << vec[i] << "<<" << i; 
      if (i == vec.size() - 1) std :: cout << std :: endl; 
      else std :: cout << " "; 
      }
   //
   //依存句法分析接口
   //parse
   void * engine3 = parser_create_parser("/Users/iceet/Mine/Bill/tools/xls/ltp_data/parser.model"); 
   vector < int > heads; 
   vector < std :: string > deprels; 
   vector < pair < int, string > > parser; 
   parser_parse(engine3, words, tags, heads, deprels); 
   for (int i = 0; i < heads.size(); ++ i) {
      std :: cout << words[i] << "\t" << tags[i] << "\t" << heads[i] << "\t" << deprels[i] << std :: endl; 
      //parser[i].first = heads[i];
      cout << heads[i] << endl; 
      // int parentIdx = atoi( heads[i].c_str() );
      //parser[i].second = deprels[i];
      parser.push_back(make_pair(static_cast < int > (heads[i]), 
      deprels[i])); 
      }
   SRL_LoadResource("/Users/iceet/Mine/Bill/tools/xls/ltp_data/srl/"); 
   vector < pair < int, vector < pair < const char * , pair < int, int > > > > > vecSRLResult; 
   SRL(words, tags, vec, parser, vecSRLResult); 
    //srl 结果，这里为什么为0呢
   cout << vecSRLResult.size() << endl; 
   int j = 0; 
   for (; j < vecSRLResult.size(); ++j) {
      vector < string > vecType; 
      vector < pair < int, 
      int > > vecBegEnd; 
      int k = 0; 
      for (; k < vecSRLResult[j].second.size(); ++k) {
         // vecType.push_back(vecSRLResult[j].second[k].first);
         //vecBegEnd.push_back(vecSRLResult[j].second[k].second);
         // std::cout << vecSRLResult[j].second[k].first[0] << "\t" <<endl;
         std :: cout << k << endl; 
         }
      cout << "--" << endl; 
      }
   return scope.Close(Undefined()); 
   }
void init(Handle < Object > exports) {
   exports -> Set(String :: NewSymbol("analyze"), 
   FunctionTemplate :: New(Method) -> GetFunction()); 
   }
NODE_MODULE(BillNLP, init)

目前中英文混合文本应该非常常见了，不知道LTP有没有考虑去实现这块。

doc(二进制)格式文档无法追踪更改 不适合在github和开源项目中使用 建议使用TeX或Markdown等重新制作

我是用cpp写的nodejs扩展，然后我这个编译透过了。在调用的时候提示这样。。
帮我分析下是什么原因呢。。
THX

node t.js
[TRACE] 2013/09/24 17:35:03 Loading segmentor model from "ltp_data/cws.model" ...
[TRACE] 2013/09/24 17:35:03 segmentor model is loaded.
[TRACE] 2013/09/24 17:35:03 Loading postagger model from "ltp_data/pos.model" ...
dyld: lazy symbol binding failed: Symbol not found: __Z26postagger_create_postaggerPKc
Referenced from: /Users/iceet/Mine/ltp/src/node/build/Release/hello.node
Expected in: dynamic lookup

dyld: Symbol not found: __Z26postagger_create_postaggerPKc
Referenced from: /Users/iceet/Mine/ltp/src/node/build/Release/hello.node
Expected in: dynamic lookup

我尝试用如下代码解密例句“我们都是**人。”

    //依存句法分析接口
    //parse
    void * engine3 = parser_create_parser("/Users/iceet/Mine/Bill/tools/xls/ltp_data/parser.model");

    vector<int>            heads;
    vector<std::string>    deprels;
    vector< pair<int, string> > parse;

    parser_parse(engine3, words, postags, heads, deprels);
    for (int i = 0; i < heads.size(); ++ i) {
        //std::cout << words[i] << "\t" << tags[i] << "\t" 
        //    << heads[i] << "\t" << deprels[i] << std::endl;
        //parser[i].first = heads[i];
       // heads[i] = heads[i];
        cout << heads[i] << deprels[i] << endl;
//         int parentIdx = atoi( heads[i].c_str() );

        //parser[i].second = deprels[i];
        parse.push_back(make_pair(heads[i],deprels[i]));
    }

期望得到结果A：

我们/r    2 SBV
都/d   2 ADV
是/v   -1 HED
**/ns   4 ATT
人/n   2 VOB
。/wp  2 WP

实际上得到的结果B：

我们/r    3 SBV
都/d   3 ADV
是/v   0 HED
**/ns   5 ATT
人/n   3 VOB
。/wp  3 WP

发现在 parser_dll.cpp 的 62 行

 int len = inst->size();
        heads.resize(len - 1);
        deprels.resize(len - 1);
        for (int i = 1; i < len; ++ i) {
            heads[i - 1] = inst->predicted_heads[i];//这里是否应该减去1？
            deprels[i - 1] = ltp::parser::Parser::model->deprels.at(
                    inst->predicted_deprelsidx[i]);
        }

在ltp中如何启用这个 wsd参数呢。。
现在的返回结果里面没有这个参数了

RT

Scanning dependencies of target postagger
[ 40%] Building CXX object src/_svmtagger/CMakeFiles/postagger.dir/dict.cpp.o
In file included from /usr/include/sys/signal.h:148,
from /usr/include/sys/wait.h:116,
from /usr/include/stdlib.h:65,
from /usr/include/c++/4.2.1/cstdlib:72,
from /usr/include/c++/4.2.1/bits/stl_algobase.h:68,
from /usr/include/c++/4.2.1/bits/char_traits.h:46,
from /usr/include/c++/4.2.1/string:47,
from /usr/local/include/boost/regex/v4/cregex.hpp:207,
from /usr/local/include/boost/cregex.hpp:27,
from /Users/wanxiang/Documents/workspace/ltp/src/_svmtagger/er.h:24,
from /Users/wanxiang/Documents/workspace/ltp/src/_svmtagger/dict.cpp:27:
/usr/include/sys/_structs.h:218: error: conflicting declaration ‘typedef struct __darwin_sigaltstack stack_t’
/Users/wanxiang/Documents/workspace/ltp/src/svmtagger/swindow.h:44: error: ‘struct stack_t’ has a previous declaration as ‘struct stack_t’
/Users/wanxiang/Documents/workspace/ltp/src/svmtagger/dict.cpp: In member function ‘void dictionary::dictWrite(char)’:
/Users/wanxiang/Documents/workspace/ltp/src/svmtagger/dict.cpp:136: warning: deprecated conversion from string constant to ‘char’
/Users/wanxiang/Documents/workspace/ltp/src/svmtagger/dict.cpp: In member function ‘void dictionary::dictCreate(FILE, int, int)’:
/Users/wanxiang/Documents/workspace/ltp/src/svmtagger/dict.cpp:250: warning: deprecated conversion from string constant to ‘char’
/Users/wanxiang/Documents/workspace/ltp/src/svmtagger/dict.cpp: In member function ‘void dictionary::dictRepairFromFile(char)’:
/Users/wanxiang/Documents/workspace/ltp/src/svmtagger/dict.cpp:325: warning: deprecated conversion from string constant to ‘char’
/Users/wanxiang/Documents/workspace/ltp/src/svmtagger/dict.cpp: In member function ‘void dictionary::dictAddBackup(char)’:
/Users/wanxiang/Documents/workspace/ltp/src/svmtagger/dict.cpp:462: warning: deprecated conversion from string constant to ‘char’
/Users/wanxiang/Documents/workspace/ltp/src/svmtagger/dict.cpp: In constructor ‘dictionary::dictionary(char, char)’:
/Users/wanxiang/Documents/workspace/ltp/src/svmtagger/dict.cpp:582: warning: deprecated conversion from string constant to ‘char’
/Users/wanxiang/Documents/workspace/ltp/src/svmtagger/dict.cpp:586: warning: deprecated conversion from string constant to ‘char’
/Users/wanxiang/Documents/workspace/ltp/src/svmtagger/dict.cpp: In constructor ‘dictionary::dictionary(char)’:
/Users/wanxiang/Documents/workspace/ltp/src/svmtagger/dict.cpp:593: warning: deprecated conversion from string constant to ‘char’
/Users/wanxiang/Documents/workspace/ltp/src/svmtagger/dict.cpp: In constructor ‘dictionary::dictionary(char, int, int)’:
/Users/wanxiang/Documents/workspace/ltp/src/svmtagger/dict.cpp:601: warning: deprecated conversion from string constant to ‘char’
make[3]: *** [src/_svmtagger/CMakeFiles/postagger.dir/dict.cpp.o] Error 1
make[2]: *** [src/_svmtagger/CMakeFiles/postagger.dir/all] Error 2
make[1]: *** [all] Error 2
make: *** [all] Error 2

现在实现的逻辑是同时将所有模块的模型load进来。

这个设计并不是很美观，考虑到有些用户只想做分词，但是却连parser的模型也加载进来。

在配置文件中指定加载的模型以及要做的任务。

ltp-3.1.0和github上最新的co试过了都有这个问题

$ bin/ltp_test ltp_data/ltp.cnf dp 1.txt 
[TRACE] 2014/04/03 12:02:18 Loading segmentor model from "ltp_data/cws.model" ...
[TRACE] 2014/04/03 12:02:18 segmentor model is loaded.
[WARNING] 2014/04/03 12:02:18 No "postagger-model" config is found
[TRACE] 2014/04/03 12:02:18 Loading parser resource from "ltp_data/parser.model"
1792
-1
[ERROR] 2014/04/03 12:02:18 /home/feng/fun/ltp/src/__ltp_dll/LTPResource.cpp: line 183: LoadParserResource(): Failed to create parser
[ERROR] 2014/04/03 12:02:18 /home/feng/fun/ltp/src/__ltp_dll/Ltp.cpp: line 128: ReadConfFile(): in LTP::parser, failed to load parser resource
Failed to load LTP
[TRACE] 2014/04/03 12:02:18 segmentor model is released.

步进调试发现是feat_opt.use_sibling被置位了，但是parser.model里面却只有一个collections

$ xxd parser.model |grep collections
0000700: 636f 6c6c 6563 7469 6f6e 7300 0000 0000  collections.....

http://q.weibo.com/849045 欢迎大家讨论有关于LTP的问题或者提出建议和意见。

相应的Traiva CI也有问

系统 windows 7，
visual studio 2008完全安装，
cmake 2.8.11.2，
报错如下：

CMake Error at thirdparty/maxent/CMakeLists.txt:80 (add_executable):
  Cannot find source file:

    getopt.c

  Tried extensions .c .C .c++ .cc .cpp .cxx .m .M .mm .h .hh .h++ .hm .hpp
  .hxx .in .txx

Wanxiangs-MacBook-Pro:ltp wanxiang$ make
[ 10%] Built target crfpp
[ 17%] Built target maxent
[ 21%] Built target tinyxml
[ 22%] Building CXX object src/__util/CMakeFiles/util.dir/Logger.cpp.o
In file included from /Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.cpp:10:
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.h:71: error: ‘MAX_PATH’ was not declared in this scope
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.h:76: error: ‘semaphore’ does not name a type
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.h:77: error: ‘semaphore’ has not been declared
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.h:78: error: ‘semaphore’ has not been declared
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.h:79: error: ‘semaphore’ has not been declared
/Users/wanxiang/Documents/workspace/ltp/src/_util/Logger.cpp: In constructor ‘CLogger::CLogger(const char, int)’:
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.cpp:16: error: ‘m_csLogger’ was not declared in this scope
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.cpp: At global scope:
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.cpp:32: error: variable or field ‘InitializeCriticalSection’ declared void
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.cpp:32: error: ‘semaphore’ was not declared in this scope
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.cpp:32: error: ‘s’ was not declared in this scope
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.cpp:42: error: variable or field ‘EnterCriticalSection’ declared void
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.cpp:42: error: ‘semaphore’ was not declared in this scope
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.cpp:42: error: ‘s’ was not declared in this scope
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.cpp:66: error: variable or field ‘LeaveCriticalSection’ declared void
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.cpp:66: error: ‘semaphore’ was not declared in this scope
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.cpp:66: error: ‘s’ was not declared in this scope
/Users/wanxiang/Documents/workspace/ltp/src/_util/Logger.cpp: In member function ‘void CLogger::Log(int, const char, _va_list_tag)’:
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.cpp:99: error: ‘m_csLogger’ was not declared in this scope
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.cpp:108: error: ‘m_OutputBuf’ was not declared in this scope
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.cpp:141: error: ‘m_OutputBuf’ was not declared in this scope
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.cpp:158: error: ‘m_OutputBuf’ was not declared in this scope
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.cpp:176: error: ‘m_OutputBuf’ was not declared in this scope
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.cpp:190: error: ‘m_OutputBuf’ was not declared in this scope
/Users/wanxiang/Documents/workspace/ltp/src/__util/Logger.h: At global scope:
/Users/wanxiang/Documents/workspace/ltp/src/_util/Logger.h:77: warning: inline function ‘static void CLogger::InitializeCriticalSection(int)’ used but never defined
/Users/wanxiang/Documents/workspace/ltp/src/_util/Logger.h:78: warning: inline function ‘static void CLogger::EnterCriticalSection(int)’ used but never defined
/Users/wanxiang/Documents/workspace/ltp/src/_util/Logger.h:79: warning: inline function ‘static void CLogger::LeaveCriticalSection(int)’ used but never defined
make[3]: *** [src/__util/CMakeFiles/util.dir/Logger.cpp.o] Error 1
make[2]: *** [src/__util/CMakeFiles/util.dir/all] Error 2
make[1]: *** [all] Error 2
make: *** [all] Error 2

c# 接口
https://github.com/yajiedesign/ltpcsharp

整理相关文档，撰写Wiki等。

正常运行多线程ltp_server不会有问题，因为理论上讲server不应该停机。但是在multi_cws_cmdline上会出现写在segmentor::rulebase里面的两个正则表达式析构时出core。

在PoSTagging的结果中含有“z”词性（依照北大标注规范），但是在Parser的训练数据中没有"z"词性。

一个解决方法是用自动词性+god dep-relation训练一个parser model。

另外，是否需要z词性还是需要再讨论。

编译成功后。执行测试：
./bin/ltp_test "ws" "test_data/test_gb.txt"

有错误：
terminate called after throwing an instance of 'std::string'

请把 _svmtagger/weight.cpp里面的相应处改为如下，否则mac编译会报 can not delete char[100]

class weight_struct_t {
    public:
        char *key = new char[100];
        hash_t *hash;

@Oneplus 帮忙看看
我复制的你们的代码,在linux下命名为seg.cc

#include <iostream>
#include <string>
#include "segment_dll.h"

int main(int argc, char * argv[]) {
    if (argc < 2) {
        std::cerr << "cws [model path]" << std::endl;
        return 1;
    }

    void * engine = segmentor_create_segmentor(argv[1]);
    if (!engine) {
        return -1;
    }
    std::vector<std::string> words;
    int len = segmentor_segment(engine, 
            "爱上一匹野马，可我的家里没有草原。", words);
    for (int i = 0; i < len; ++ i) {
        std::cout << words[i] << "|";
    }
    std::cout << std::endl;
    segmentor_release_segmentor(engine);
    return 0;
}

然后执行
g++ seg.cc segmentor.a -o seg
好像编译通不过(ps:我对linux下c++编程不熟悉)

segmentor.cpp:(.text._ZN5boost9re_detail12perl_matcherIN9__gnu_cxx17__normal_iteratorIPKcSsEESaINS_9sub_matchIS6_EEENS_12regex_traitsIcNS_16cpp_regex_traitsIcEEEEE8find_impEv[boost::re_detail::perl_matcher<__gnu_cxx::__normal_iterator<char const*, std::basic_string<char, std::char_traits<char>, std::allocator<char> > >, std::allocator<boost::sub_match<__gnu_cxx::__normal_iterator<char const*, std::basic_string<char, std::char_traits<char>, std::allocator<char> > > > >, boost::regex_traits<char, boost::cpp_regex_traits<char> > >::find_imp()]+0x18f): undefined reference to `boost::re_detail::put_mem_block(void*)'
segmentor.cpp:(.text._ZN5boost9re_detail12perl_matcherIN9__gnu_cxx17__normal_iteratorIPKcSsEESaINS_9sub_matchIS6_EEENS_12regex_traitsIcNS_16cpp_regex_traitsIcEEEEE8find_impEv[boost::re_detail::perl_matcher<__gnu_cxx::__normal_iterator<char const*, std::basic_string<char, std::char_traits<char>, std::allocator<char> > >, std::allocator<boost::sub_match<__gnu_cxx::__normal_iterator<char const*, std::basic_string<char, std::char_traits<char>, std::allocator<char> > > > >, boost::regex_traits<char, boost::cpp_regex_traits<char> > >::find_imp()]+0x361): undefined reference to `boost::re_detail::put_mem_block(void*)'
collect2: ld returned 1 exit status

As I know, main reason for compiling error is the wrongly defined macro

ifdef __WIN32__

these kind of macro is designed for MSVC, however mingw under windows also triggered this macro.

another issue is the usage of hashmap under mingw is still a mystery to me.

Hi all,

The srl module has been updated, as well as a new maxent package. Some details are shown below:

• A predicate recognition (PRG) module is added, displacing the previous POS-based recognizer.
• A new maxent pack is added, placed at the thirdparty/ folder. Three solvers: L1-OWLQN, L1-SGD and L2-LBFGS are provided.
• The latest SRL module by default uses the L1-OWLQN solver, in order to obtain a sparse and thus much smaller model.
• An integral training and testing suite for PRG/SRL is provided (see src/_srl/lgsrl.cpp), displacing the previous mass of EXEs. Besides, several configuration files for srl training are updated (see tools/train/assets/)
• I did not reorganize the source architecture, such as renaming _srl to srl. Yijia please help do it.

请问与3.0.0 alpha 相匹配的最新模型文件，哪里可以下载到？？

系统是ubuntu12.04

Input sentence is: **你好？
-->>debug make xml
[XML4NLP ERROR REPORT]
description : Error document empty.
location :
row : 0
col : 0

===================

[ERROR] 2013/09/02 08:51:15 /home/parallels/ltp/src/__ltp_dll/Ltp.cpp: line 147: splitSentence_dummy(): in LTP::splitsent, There is no paragraph in doc,
[ERROR] 2013/09/02 08:51:15 /home/parallels/ltp/src/__ltp_dll/Ltp.cpp: line 148: splitSentence_dummy(): you may have loaded a blank file or have not loaded a file yet.
[ERROR] 2013/09/02 08:51:15 /home/parallels/ltp/src/__ltp_dll/Ltp.cpp: line 182: wordseg(): in LTP::wordseg, failed to perform split sentence preprocess.
[ERROR] 2013/09/02 08:51:15 /home/parallels/ltp/src/__ltp_dll/Ltp.cpp: line 233: postag(): in LTP::postag, failed to perform word segment preprocess
[ERROR] 2013/09/02 08:51:15 /home/parallels/ltp/src/__ltp_dll/Ltp.cpp: line 284: ner(): in LTP::ner, failed to perform postag preprocess
[ERROR] 2013/09/02 08:51:15 /home/parallels/ltp/src/__ltp_dll/Ltp.cpp: line 397: srl(): in LTP::srl, failed to perform ner preprocess
Result is: HTTP/1.1 200 OK

【视频日媒称越南因南海争端停播**央视节目越南停播央视日媒新浪视频】

最后一个】会标错

Ⅰ、大豆及其制品。

Ⅰ会标错

应该加入一个特殊字符识别的特征。

我下载了新版本的源码然后编译：

[ 34%] Building CXX object src/segmentor/CMakeFiles/otcws.dir/otcws.cpp.o
In file included from 。。。/LTP/src/segmentor/otcws.cpp:2:
。。。/LTP/src/utils/cfgparser.hpp:11:10: fatal error:
'tr1/unordered_map' file not found

include <tr1/unordered_map>

     ^

1 error generated.
make[3]: *** [src/segmentor/CMakeFiles/otcws.dir/otcws.cpp.o] Error 1
make[2]: *** [src/segmentor/CMakeFiles/otcws.dir/all] Error 2
make[1]: *** [all] Error 2
make: *** [all] Error 2

现在LTP 完整运行需要的内存大概在1400M-2000M左右，，
这个有考虑把内存降低的打算么？

已经用不上了。

src/segmentor/segment_dll.cpp line 67-71

在根目录下的CMakeLists.txt会在win32环境下排斥编译unittest和ltp_server，但是在win64下仍旧会编译这两个模块。

字母和数字混合时，错误的将字母和数字分隔开，如“惠普d2015打印机”中，“的”和“2015”分成了两个词。

RT

您好，文档中的的模型没有下载链接，能给个链接吗？
谢谢！

there should be a CXX_FLAGS -DHAVE_CONFIG_H set in thirdparty/maxent/CMakeFiles.

在mac和linux上编译出了ltp_test2，和ltp_test2，还有ltp_test_xml，这3个可执行程序怎么用，另外linux上有python的接口吗？还是必须自己写程序调用链接库的函数?

在这个行数里面的： SRLBaselineExt::ExtractPrgFeatures(vector< vector >& vecPrgFeatures) 的83行。。

我的调用如下：

SRL_LoadResource("/Users/iceet/Mine/Bill/tools/xls/ltp_data/srl/");

    vector< pair< int, vector< pair<const char *, pair< int, int > > > > > vecSRLResult;

    std::cout << "==ing===" << endl;

    SRL(words, tags, vec, parser, vecSRLResult);//这里执行的时候报错了

    std::cout << "=======e" << std::endl;

最终定位到代码 SRLBaselineExt.cpp

    for (size_t row = 1; row <= row_count; ++row)
    {
        vector<string> instance;
        for (size_t i = 0; i < m_prgFeatureNumbers.size(); ++i)
        {
            string feature = m_prgFeaturePrefixes[i] + "@"
                + vec_feature_values[i][row]; //这里有问题
            instance.push_back(feature);
        }
        vecPrgFeatures.push_back(instance);
    }

谢谢：
文档中，没有
https://github.com/HIT-SCIR/ltp/blob/master/doc/ltp-document-3.0.md#%E8%AF%AD%E4%B9%89%E8%A7%92%E8%89%B2%E6%A0%87%E6%B3%A8%E6%8E%A5%E5%8F%A3

ltp::utility::Template是整个ltp里面最常用的基础数据结构，提供从特征模板中实例化特征字符串的功能。例如，特征模板T=3={w0}-{p0}。在w0=am，p0=v的情况下被实例化为3=am-v。

特征模板的实现方法是将特征模板拆解成若干tokens(3=, w0, -, p0)，将这些tokens存储在Template_Token_Cache的单件里。然后将Template转换为token对应的index。在这个例子里，模板T被表示为(0,1,2,3)的index列表。

每次实例化前需要定义一个Template::Data的类型，用以存储每个实例化的token。Template::Data对于Token_Cache建立一个副本，然后在调用set方法时将相应key实例化为value。render模板的过程变成将模板里的index列表对应的token拼接的过程。

旧版的实现在设置key的时候需要一步查找。由于认为token的个数比较少，这一过程直接使用线性查找。最近的unittest中发现这一过程可以采用hashmap替换从而获得更快的速度。

这一修改将会直接影响各模块的速度（训练，解码）。从初步实验来看，这一修改在二阶sibling模型上能够获得40%的速度提升，使得分析速度从7.35句/s提高到12.7句/s（因测试机不同而有差异）。还需进一步进行测试。

句子：你好勤快
理想分词结果:你 好 勤快
实际分词结果：你好 勤快