├── .gitignore ├── LICENSE ├── README.md ├── TreeIndex.cpp ├── TreeIndex.h ├── en.txt ├── main.cpp ├── noderank.png └── treesearch.png /.gitignore: -------------------------------------------------------------------------------- 1 | # Compiled Object files 2 | *.slo 3 | *.lo 4 | *.o 5 | *.obj 6 | 7 | # Precompiled Headers 8 | *.gch 9 | *.pch 10 | 11 | # Compiled Dynamic libraries 12 | *.so 13 | *.dylib 14 | *.dll 15 | 16 | # Fortran module files 17 | *.mod 18 | 19 | # Compiled Static libraries 20 | *.lai 21 | *.la 22 | *.a 23 | *.lib 24 | 25 | # Executables 26 | *.exe 27 | *.out 28 | *.app 29 | -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | GNU LESSER GENERAL PUBLIC LICENSE 2 | Version 3, 29 June 2007 3 | 4 | Copyright (C) 2007 Free Software Foundation, Inc. 5 | Everyone is permitted to copy and distribute verbatim copies 6 | of this license document, but changing it is not allowed. 7 | 8 | 9 | This version of the GNU Lesser General Public License incorporates 10 | the terms and conditions of version 3 of the GNU General Public 11 | License, supplemented by the additional permissions listed below. 12 | 13 | 0. Additional Definitions. 14 | 15 | As used herein, "this License" refers to version 3 of the GNU Lesser 16 | General Public License, and the "GNU GPL" refers to version 3 of the GNU 17 | General Public License. 18 | 19 | "The Library" refers to a covered work governed by this License, 20 | other than an Application or a Combined Work as defined below. 21 | 22 | An "Application" is any work that makes use of an interface provided 23 | by the Library, but which is not otherwise based on the Library. 24 | Defining a subclass of a class defined by the Library is deemed a mode 25 | of using an interface provided by the Library. 26 | 27 | A "Combined Work" is a work produced by combining or linking an 28 | Application with the Library. The particular version of the Library 29 | with which the Combined Work was made is also called the "Linked 30 | Version". 31 | 32 | The "Minimal Corresponding Source" for a Combined Work means the 33 | Corresponding Source for the Combined Work, excluding any source code 34 | for portions of the Combined Work that, considered in isolation, are 35 | based on the Application, and not on the Linked Version. 36 | 37 | The "Corresponding Application Code" for a Combined Work means the 38 | object code and/or source code for the Application, including any data 39 | and utility programs needed for reproducing the Combined Work from the 40 | Application, but excluding the System Libraries of the Combined Work. 41 | 42 | 1. Exception to Section 3 of the GNU GPL. 43 | 44 | You may convey a covered work under sections 3 and 4 of this License 45 | without being bound by section 3 of the GNU GPL. 46 | 47 | 2. Conveying Modified Versions. 48 | 49 | If you modify a copy of the Library, and, in your modifications, a 50 | facility refers to a function or data to be supplied by an Application 51 | that uses the facility (other than as an argument passed when the 52 | facility is invoked), then you may convey a copy of the modified 53 | version: 54 | 55 | a) under this License, provided that you make a good faith effort to 56 | ensure that, in the event an Application does not supply the 57 | function or data, the facility still operates, and performs 58 | whatever part of its purpose remains meaningful, or 59 | 60 | b) under the GNU GPL, with none of the additional permissions of 61 | this License applicable to that copy. 62 | 63 | 3. Object Code Incorporating Material from Library Header Files. 64 | 65 | The object code form of an Application may incorporate material from 66 | a header file that is part of the Library. You may convey such object 67 | code under terms of your choice, provided that, if the incorporated 68 | material is not limited to numerical parameters, data structure 69 | layouts and accessors, or small macros, inline functions and templates 70 | (ten or fewer lines in length), you do both of the following: 71 | 72 | a) Give prominent notice with each copy of the object code that the 73 | Library is used in it and that the Library and its use are 74 | covered by this License. 75 | 76 | b) Accompany the object code with a copy of the GNU GPL and this license 77 | document. 78 | 79 | 4. Combined Works. 80 | 81 | You may convey a Combined Work under terms of your choice that, 82 | taken together, effectively do not restrict modification of the 83 | portions of the Library contained in the Combined Work and reverse 84 | engineering for debugging such modifications, if you also do each of 85 | the following: 86 | 87 | a) Give prominent notice with each copy of the Combined Work that 88 | the Library is used in it and that the Library and its use are 89 | covered by this License. 90 | 91 | b) Accompany the Combined Work with a copy of the GNU GPL and this license 92 | document. 93 | 94 | c) For a Combined Work that displays copyright notices during 95 | execution, include the copyright notice for the Library among 96 | these notices, as well as a reference directing the user to the 97 | copies of the GNU GPL and this license document. 98 | 99 | d) Do one of the following: 100 | 101 | 0) Convey the Minimal Corresponding Source under the terms of this 102 | License, and the Corresponding Application Code in a form 103 | suitable for, and under terms that permit, the user to 104 | recombine or relink the Application with a modified version of 105 | the Linked Version to produce a modified Combined Work, in the 106 | manner specified by section 6 of the GNU GPL for conveying 107 | Corresponding Source. 108 | 109 | 1) Use a suitable shared library mechanism for linking with the 110 | Library. A suitable mechanism is one that (a) uses at run time 111 | a copy of the Library already present on the user's computer 112 | system, and (b) will operate properly with a modified version 113 | of the Library that is interface-compatible with the Linked 114 | Version. 115 | 116 | e) Provide Installation Information, but only if you would otherwise 117 | be required to provide such information under section 6 of the 118 | GNU GPL, and only to the extent that such information is 119 | necessary to install and execute a modified version of the 120 | Combined Work produced by recombining or relinking the 121 | Application with a modified version of the Linked Version. (If 122 | you use option 4d0, the Installation Information must accompany 123 | the Minimal Corresponding Source and Corresponding Application 124 | Code. If you use option 4d1, you must provide the Installation 125 | Information in the manner specified by section 6 of the GNU GPL 126 | for conveying Corresponding Source.) 127 | 128 | 5. Combined Libraries. 129 | 130 | You may place library facilities that are a work based on the 131 | Library side by side in a single library together with other library 132 | facilities that are not Applications and are not covered by this 133 | License, and convey such a combined library under terms of your 134 | choice, if you do both of the following: 135 | 136 | a) Accompany the combined library with a copy of the same work based 137 | on the Library, uncombined with any other library facilities, 138 | conveyed under the terms of this License. 139 | 140 | b) Give prominent notice with the combined library that part of it 141 | is a work based on the Library, and explaining where to find the 142 | accompanying uncombined form of the same work. 143 | 144 | 6. Revised Versions of the GNU Lesser General Public License. 145 | 146 | The Free Software Foundation may publish revised and/or new versions 147 | of the GNU Lesser General Public License from time to time. Such new 148 | versions will be similar in spirit to the present version, but may 149 | differ in detail to address new problems or concerns. 150 | 151 | Each version is given a distinguishing version number. If the 152 | Library as you received it specifies that a certain numbered version 153 | of the GNU Lesser General Public License "or any later version" 154 | applies to it, you have the option of following the terms and 155 | conditions either of that published version or of any later version 156 | published by the Free Software Foundation. If the Library as you 157 | received it does not specify a version number of the GNU Lesser 158 | General Public License, you may choose any version of the GNU Lesser 159 | General Public License ever published by the Free Software Foundation. 160 | 161 | If the Library as you received it specifies that a proxy can decide 162 | whether future versions of the GNU Lesser General Public License shall 163 | apply, that proxy's public statement of acceptance of any version is 164 | permanent authorization for you to choose that version for the 165 | Library. 166 | 167 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # Eli-Index : Ranked Prefix Search for Large Data on External Memory 2 | Previously I wrote (https://github.com/maksimpiriyev/Search-As-You-Type-Index-master) about fast prefix search for ~10,000 key-value pairs designed for mobile.Now,Eli-Index designed to handle more than 1,000,000 key-value pairs,make fast prefix search on them,and furthermore bring the most ranked one from index.It stores index and data at the same time on external memory not in RAM (caches the nodes while it needs) 3 | 4 | 5 | Most of the prefix search indices are designed for server side use,they store index in the RAM so that has intialization latency which prenvents opening application more faster in mobile(using threads also slows down the main thread).Eli-Index is designed to overcome this problem. 6 | Eli-Index is designed over AVL Tree,Segmented Tree and with Heap.Its nodes has the features of both AVL Tree and Segmented Tree.It makes prefix search over AVL Tree ,and gets the result list in sorted order while iterating on the nodes of segmented tree with the help of heap. It might also be the first AVL Tree implemented for string prefix search as I couldn't find any paper or implementation that use AVL Tree for string Prefix Search. Eli-Index has preferred to use AVL Tree instead of Trie Index because if Trie Index is implemented in External Memory,Either it is fast and uses more memory or it is slower because of more random file access. 7 | 8 | ### Class structure: 9 | ```c++ 10 | class TreeIndex{ 11 | 12 | public: 13 | TreeIndex(const string fileName); 14 | void save(); 15 | long size(); 16 | vector > search(wstring txt, int maxCount = 100); 17 | vector searchOnlyString(wstring txt, int maxCount = 100); 18 | tuple* find(wstring txt); 19 | void add(wstring txt, long statCount = 1/*rank*/,long value = 0); 20 | void loadNgram(string fileName, int ngram); 21 | void loadTopWords(string fileName,int maxCount=-1); 22 | bool isExist(wstring s); 23 | void setText(Node* n, wstring txt); 24 | void clearCache(); 25 | int cacheSize(); 26 | int treeHeight(); 27 | }; 28 | ``` 29 | Its usage : 30 | ```c++ 31 | TreeIndex xx("index_en.tf"); 32 | cout << xx.treeHeight() << endl; 33 | xx.add(L"pretend"); 34 | xx.add(L"predict",1); 35 | xx.add(L"prediction",2,1); 36 | auto xff = xx.search(L"pre",100); 37 | for (size_t j = 0; j < xff.size(); j++) 38 | { 39 | wcout << std::get<1>(xff[j]) << endl; 40 | } 41 | 42 | ``` 43 | 44 | ### Prefix Search with AVL Tree 45 | Note that AVL Tree is the same as it was with same operations,but Prefix Search on AVL Tree is little bit different than the regular search on AVL Tree.Lets say we are searching the prefix P.We always compare the prefix of the text in the node with length |P|. 46 | 47 | > * If node N is greater than the prefix then its right nodes do not contain that prefix.
48 | > * If node N is smaller than the prefix then its left nodes do not contain that prefix.
49 | > * The case that left node is smaller than P and right node is greater than P might exists only at the node itself or (either in left subtree or in right subtree),but not left and right subtree at the same time.This is why the complexity is logN not 2^N 50 | ![alt tag](https://raw.githubusercontent.com/maksimpiriyev/Eli-Index/master/treesearch.png) 51 | 52 | 53 | ### Top K Ranks with Segmented Tree 54 | Segmented Tree is well known structure to find the min-max values in any tree. Eli-Index iterates between top min-max values using heap. 55 | 56 | > * Adding (K+1)th element to the heap and removing the least element guarantees that heap has the top K elements. 57 | ![alt tag](https://raw.githubusercontent.com/maksimpiriyev/Eli-Index/master/noderank.png) 58 | 59 | ### Insert Comlexity 60 | Insert complexity is as same as the AVL tree O(logN) 61 | 62 | ### Search Complexity for top K words 63 | Segmented Tree has the O(logN) complexity for finding the node with maximum rank value,using heap and finding next K-1 complexity brings extra O(lgK*lglgN) overhead. 64 | > Total Complexity: O(K* lgK* lgN* lglgN) . 65 | 66 | ### Compilation 67 | It is written in C++11. 68 | > g++ -Ofast -std=c++11 -o prog main.cpp TreeFile.cpp 69 | 70 | ### Eli-Index vs SQL 71 | Note that nowadays most of the databases have prefix search with their full-text search functionality.Lets say you have more than 1,000,000 words and 100,000 of them starts with "a*" and you want search all of them ordered by their rank.This time Eli-Index and SQL will have very similar performance results.But If you search top 10 ranked(user-defined) results from these 100,000 words then Eli-Index will drammatically beat SQL,because SQL is abstract it retrieves and then sorts for column and gets top results, instead El-Index finds top nodes first by default 72 | 73 | ### OS Cache and Initial Latency 74 | Some Operating Systems has the policy to cache whole file at first acess before you do any operation on it (OS-wide first access),and it brings extra latency in the first operation,but meanwhile makes faster the other operations.But, you can use fadvise,fcntrl etc to disable cache or tell the read pattern is random. 75 | 76 | ### Apps that use this index 77 | Typing auto-complete for mac : http://elithetyper.com 78 | 79 | by Maksim Piriyev 80 | -------------------------------------------------------------------------------- /TreeIndex.cpp: -------------------------------------------------------------------------------- 1 | #include 2 | #include 3 | #include 4 | #include 5 | #include 6 | #include 7 | 8 | #include 9 | #include "TreeIndex.h" 10 | #include 11 | #include 12 | #include 13 | using namespace std; 14 | #define MAXNODECOUNT 40 15 | #define COUNTTHRESHOLD 4000000 16 | 17 | static map dictCharToIndex = { { ' ', 0 }, { 'a', 1 }, { 'b', 2 }, { 'c', 3 }, { 'd', 4 }, { 'e', 5 }, { 'f', 6 }, { 'g', 7 }, { 'h', 8 }, { 'i', 9 }, 18 | { 'j', 10 }, { 'k', 11 }, { 'l', 12 }, { 'm', 13 }, { 'n', 14 }, { 'o', 15 }, { 'p', 16 }, { 'q', 17 }, { 'r', 18 }, { 's', 19 }, 19 | { 't', 20 }, { 'u', 21 }, { 'v', 22 }, { 'w', 23 }, { 'x', 24 }, { 'y', 25 }, { 'z', 26 }, { '.', 27 }, { '-', 28 }, { 's', 29 } 20 | 21 | }; 22 | static map dictIndexToChar = { { 0,' ' }, { 1,'a' }, { 2,'b' }, { 3,'c'}, { 4,'d' }, { 5,'e' }, { 6,'f' }, { 7,'g' }, {8, 'h' }, { 9,'i' }, 23 | { 10,'j' }, { 11,'k' }, { 12,'l' }, { 13,'m' }, { 14,'n'}, { 15,'o' }, { 16,'p' }, { 17,'q' }, { 18,'r'}, { 19,'s' }, 24 | { 20,'t' }, { 21,'u' }, { 22,'v' }, { 23,'w' }, { 24,'x'}, { 25,'y' }, { 26,'z' }, { 27,'.' }, { 28,'-'}, { 29,'s' } 25 | 26 | }; 27 | 28 | namespace Index{ 29 | 30 | typedef unsigned int uint; 31 | typedef unsigned short word; 32 | typedef unsigned char byte; 33 | 34 | int UTF16to8(const wchar_t* w, char* s ) { 35 | uint c; 36 | word* p = (word*)w; 37 | byte* q = (byte*)s; byte* q0 = q; 38 | while( 1 ) { 39 | c = *p++; 40 | if( c==0 ) break; 41 | if( c<0x080 ) *q++ = c; else 42 | if( c<0x800 ) *q++ = 0xC0+(c>>6), *q++ = 0x80+(c&63); else 43 | *q++ = 0xE0+(c>>12), *q++ = 0x80+((c>>6)&63), *q++ = 0x80+(c&63); 44 | } 45 | *q = 0; 46 | return q-q0; 47 | } 48 | 49 | Node::Node(){ 50 | this->statCount = this->maxStatCount = this->nodeHeight = 0; 51 | this->textIndex = -1; 52 | this->text = NULL; 53 | this->leftNodeId = this->rightNodeId = -1; 54 | this->parentTextNodeId=-1; 55 | } 56 | 57 | TreeIndex::TreeIndex(const string fileName){ 58 | this->fileName = fileName; 59 | // file=fopen( "C:\\Users\\Maksim Piriyev\\index_en.tf", "rb+"); 60 | 61 | const char* cfileName=fileName.c_str(); 62 | printf("%s",cfileName); 63 | file = fopen( cfileName, "rb+"); 64 | wordFile = fopen( (fileName + ".w").c_str(), "rb+"); 65 | // int fd= (int)file; 66 | //fcntl(F_NOCACHE) 67 | cout<file != NULL) fclose(this->file); 100 | if (this->wordFile != NULL) fclose(this->wordFile); 101 | cache.clear(); 102 | } 103 | void TreeIndex::loadTopWords(string fileName,int maxCount){ 104 | wifstream in(fileName); 105 | wstring word; 106 | int counter = 1; 107 | in.imbue(std::locale(std::locale::classic(), new std::codecvt_utf8)); 108 | 109 | //maxCount = 100000; 110 | 111 | vector lst; 112 | while (in >> word && counter<=maxCount){ 113 | lst.push_back(word); 114 | ++counter; 115 | } 116 | cout << "readed " <0 && i % 3000 == 0){ 132 | cout << i << endl; 133 | // es = ensureStats(); 134 | // 135 | // if (!es){ 136 | // cout << "problem:" << i << endl; 137 | // return; 138 | // } 139 | // save(); 140 | // clearCache(); 141 | } 142 | } 143 | save(); 144 | clearCache(); 145 | cout<<"finished"<)); 160 | 161 | int statCount; 162 | wstring words[10]; 163 | int counter = 0; 164 | vector stats; 165 | vector lst; 166 | while (in >> statCount){ 167 | wstring rslt; 168 | for (int i = 0; i < ngram; i++) 169 | { 170 | in >> words[i]; 171 | if (i>0) rslt += L" "; 172 | rslt += words[i]; 173 | } 174 | stats.push_back(statCount); 175 | lst.push_back(rslt); 176 | 177 | 178 | counter++; 179 | if (counter == 1266){ 180 | int ff = 0; 181 | cout << "AS"<add(lst[i], stats[i]); 188 | } 189 | save(); 190 | clearCache(); 191 | cout << counter << endl; 192 | cout << treeHeight() << endl; 193 | stats.clear(); 194 | lst.clear(); 195 | 196 | } 197 | } 198 | for (size_t i = 0; i < stats.size(); i++) 199 | { 200 | this->add(lst[i], stats[i]); 201 | } 202 | this->save(); 203 | this->clearCache(); 204 | return; 205 | int j = stats.size() / 2; 206 | for (size_t i = 0; (i + j) = 0; i++) 207 | { 208 | this->add(lst[j + i], stats[j + i]); 209 | if (i>0) 210 | this->add(lst[j - i], stats[j - i]); 211 | counter++; 212 | if (counter % 1000 == 0) 213 | this->save(); 214 | } 215 | 216 | } 217 | bool TreeIndex::ensureStats(int node,long stat){ 218 | if (node < 0) return true; 219 | auto n = this->get(node); 220 | if (node == 0) stat = n->maxStatCount; 221 | if (n->maxStatCount < n->statCount || stat < n->maxStatCount) return false; 222 | bool r = ensureStats(n->leftNodeId,stat); 223 | bool r2 = ensureStats(n->rightNodeId,stat); 224 | bool r4 = false; 225 | return r&&r2; 226 | return ensureStats(n->leftNodeId,stat) && ensureStats(n->rightNodeId,stat); 227 | //return true; 228 | } 229 | void TreeIndex::clearCache(){ 230 | for (auto t = cache.begin(); t != cache.end(); t++) 231 | { 232 | auto i = *t; 233 | if(i.second->text!=NULL) 234 | delete[] i.second->text; 235 | } 236 | cache.clear(); 237 | } 238 | int TreeIndex::cacheSize(){ 239 | 240 | return cache.size(); 241 | } 242 | Node* TreeIndex::get(long nodeId){ 243 | if (nodeId < 0) return NULL; 244 | if (cache.count(nodeId) > 0) return cache[nodeId]; 245 | 246 | //return NULL; 247 | long sz = this->size(); 248 | if (sz <= nodeId) return NULL; 249 | 250 | 251 | Node* n = new Node(); 252 | 253 | fseek(this->file, sizeof(Node)*nodeId, SEEK_SET); 254 | fread(n, sizeof(Node), 1, this->file); 255 | n->text = new wchar_t[n->textCount + 1]; 256 | fseek(this->wordFile, n->textIndex, SEEK_SET); 257 | fread(n->text, sizeof(wchar_t), n->textCount + 1, this->wordFile); 258 | cache[nodeId] = n; 259 | return n; 260 | } 261 | 262 | void TreeIndex::save(){ 263 | for (auto t = saveQueue.begin(); t != saveQueue.end(); t++) 264 | { 265 | long nodeId = *t; 266 | Node* n = cache[nodeId]; 267 | if (n->textIndex == -1){ 268 | fseek(this->wordFile, 0, SEEK_END); 269 | n->textIndex = ftell(wordFile); 270 | fwrite(n->text, sizeof(wchar_t), n->textCount + 1, this->wordFile); 271 | } 272 | fseek(this->file, sizeof(Node)*nodeId, SEEK_SET); 273 | fwrite(n, sizeof(Node), 1, this->file); 274 | } 275 | fflush(file); 276 | fflush(wordFile); 277 | 278 | saveQueue.clear(); 279 | // cache.clear(); 280 | } 281 | int indexOf(int ch){ 282 | 283 | return dictCharToIndex[ch]; 284 | } 285 | wchar_t charFromIndex(int i){ 286 | return dictIndexToChar[i]; 287 | } 288 | bool hasPrefix(wstring txt, wstring prefix){ 289 | if (txt.length() < prefix.length()) return false; 290 | 291 | return prefix == txt.substr(0, prefix.length()); 292 | } 293 | void TreeIndex::setText(Node* n, wstring txt){ 294 | n->text = new wchar_t[txt.length() + 1]; 295 | wcscpy(n->text, txt.c_str()); 296 | n->textCount = txt.length(); 297 | } 298 | void TreeIndex::add(wstring txt, long statCount,long value){ 299 | //txt = txt.toLower(); 300 | long nodeId = 0, subNodeId = -1; 301 | Node* n = this->get(nodeId); 302 | if (n == NULL) { 303 | nodeId = newNode(); 304 | n = this->get(nodeId); 305 | } 306 | 307 | vector l = { nodeId }; 308 | 309 | bool did = true; 310 | while (did){ 311 | did = false; 312 | if (n->text != NULL && wstring(n->text) == txt){ 313 | //if (n->statCount < COUNTTHRESHOLD) n->statCount = COUNTTHRESHOLD; 314 | n->statCount += statCount; 315 | n->value = value; 316 | saveQueue.insert(nodeId); 317 | updateMaxStatCounts(l); 318 | return; 319 | } 320 | if (n->leftNodeId > 0 && wstring(n->text) > txt){ 321 | nodeId = n->leftNodeId; 322 | n = this->get(n->leftNodeId); 323 | l.push_back(nodeId); 324 | did = true; 325 | } 326 | else if (n->rightNodeId > 0 && wstring(n->text) <= txt){ 327 | nodeId = n->rightNodeId; 328 | n = this->get(n->rightNodeId); 329 | l.push_back(nodeId); 330 | did = true; 331 | } 332 | } 333 | if (n->text == NULL){ 334 | setText(n, txt); 335 | saveQueue.insert(nodeId); 336 | } 337 | else if (n->leftNodeId < 0 && wstring(n->text) >= txt){ 338 | saveQueue.insert(nodeId); 339 | 340 | long newNodeId = newNode(); 341 | n->leftNodeId = newNodeId; 342 | n = this->get(newNodeId); 343 | setText(n, txt); 344 | l.push_back(newNodeId); 345 | } 346 | else if (n->rightNodeId < 0 && wstring(n->text) <= txt){ 347 | saveQueue.insert(nodeId); 348 | 349 | long newNodeId = newNode(); 350 | n->rightNodeId = newNodeId; 351 | n = this->get(newNodeId); 352 | setText(n, txt); 353 | l.push_back(newNodeId); 354 | 355 | } 356 | n->statCount = n->maxStatCount = statCount; 357 | n->value = value; 358 | 359 | updateMaxStatCounts(l); 360 | } 361 | int charSize(){ 362 | 363 | return dictCharToIndex.size(); 364 | } 365 | 366 | void TreeIndex::addSave(wstring txt, long statCount){ 367 | add(txt,statCount); 368 | save(); 369 | } 370 | void TreeIndex::updateMaxStatCounts(vector& nodes){ 371 | bool firsttime = true; 372 | long maxStatCount = 0; 373 | bool did = true; 374 | for (int i = nodes.size() - 1; i >= 0 && did; i--) 375 | { 376 | //did = false; 377 | Node* n = this->get(nodes[i]); 378 | if (firsttime){ 379 | firsttime = false; 380 | if (n->maxStatCount < n->statCount){ 381 | n->maxStatCount = n->statCount; 382 | saveQueue.insert(nodes[i]); 383 | } 384 | // maxStatCount = n->maxStatCount; 385 | did = true; 386 | balanceNode(nodes[i]); 387 | continue; 388 | } 389 | /*if (n->maxStatCount < maxStatCount){ 390 | n->maxStatCount = maxStatCount; 391 | saveQueue.insert(nodes[i]); 392 | did = true; 393 | }*/ 394 | //maxStatCount = max(maxStatCount, n->maxStatCount); 395 | did |= balanceNode(nodes[i]); 396 | 397 | } 398 | } 399 | //void TreeIndex::incrementUsage(wstring txt, int statCount){ 400 | // this->add(txt, statCount); 401 | //} 402 | bool TreeIndex::isExist(wstring txt){ 403 | auto r = this->search(txt); 404 | if (r.size() > 0 && std::get<1>(r[0]) == txt) return true; 405 | return false; 406 | } 407 | int TreeIndex::treeHeight(){ 408 | return nodeHeight(0); 409 | } 410 | int TreeIndex::nodeHeight(long nodeId){ 411 | Node* n = this->get(nodeId); 412 | return n == NULL ? 0 : n->nodeHeight; 413 | } 414 | int TreeIndex::updateNodeHeight(long nodeId){ 415 | Node* n = this->get(nodeId); 416 | if (n == NULL) return 0; 417 | Node* l = n->leftNodeId < 0 ? NULL : this->get(n->leftNodeId); 418 | Node* r = n->rightNodeId < 0 ? NULL : this->get(n->rightNodeId); 419 | int h = max(l == NULL ? 0 : l->nodeHeight, r == NULL ? 0 : r->nodeHeight) + 1; 420 | if (h != n->nodeHeight) saveQueue.insert(nodeId); 421 | 422 | return n->nodeHeight = h; 423 | } 424 | int TreeIndex::updateNodeStat(long nodeId){ 425 | Node* n = this->get(nodeId); 426 | if (n == NULL) return 0; 427 | Node* l = n->leftNodeId < 0 ? NULL : this->get(n->leftNodeId); 428 | Node* r = n->rightNodeId < 0 ? NULL : this->get(n->rightNodeId); 429 | long st = max(n->statCount, max(l == NULL ? n->statCount : l->maxStatCount, r == NULL ? n->statCount : r->maxStatCount)); 430 | if (st != n->maxStatCount) saveQueue.insert(nodeId); 431 | 432 | return n->maxStatCount = st; 433 | } 434 | 435 | int TreeIndex::balanceFactor(long nodeId){ 436 | Node* n = this->get(nodeId); 437 | if (n == NULL) return 0; 438 | int lh = (n->leftNodeId == 0 ? 0 : nodeHeight(n->leftNodeId)), rh = n->rightNodeId == 0 ? 0 : nodeHeight(n->rightNodeId); 439 | return lh - rh; 440 | } 441 | 442 | bool TreeIndex::balanceNode(long nodeId){ 443 | Node* n = this->get(nodeId); 444 | if (n == NULL) return false; 445 | int bf = balanceFactor(nodeId); 446 | bool did = false; 447 | if (bf <= -2){ 448 | int lbf = balanceFactor(n->rightNodeId); 449 | if (lbf >= 1) 450 | rotateRight(n->rightNodeId); 451 | rotateLeft(nodeId); 452 | did = true; 453 | } 454 | else if (bf >= 2){ 455 | int rbf = balanceFactor(n->leftNodeId); 456 | if (rbf <= -1) 457 | rotateLeft(n->leftNodeId); 458 | rotateRight(nodeId); 459 | did = true; 460 | } 461 | updateNodeHeight(nodeId); 462 | updateNodeStat(nodeId); 463 | return did; 464 | } 465 | void TreeIndex::rotateRight(long nodeId){ 466 | Node* n = this->get(nodeId); 467 | int ln = n->leftNodeId; 468 | if (n == NULL || ln < 0) return; 469 | saveQueue.insert(nodeId); 470 | saveQueue.insert(ln); 471 | 472 | Node* l = this->get(n->leftNodeId); 473 | swap(*n, *l); 474 | l->leftNodeId = n->rightNodeId; 475 | n->rightNodeId = ln; 476 | updateNodeHeight(ln); 477 | updateNodeStat(ln); 478 | updateNodeHeight(nodeId); 479 | updateNodeStat(nodeId); 480 | 481 | } 482 | 483 | void TreeIndex::rotateLeft(long nodeId){ 484 | Node* n = this->get(nodeId); 485 | int rn = n->rightNodeId; 486 | if (n == NULL || rn < 0) return; 487 | saveQueue.insert(nodeId); 488 | saveQueue.insert(rn); 489 | 490 | Node* r = this->get(n->rightNodeId); 491 | swap(*n, *r); 492 | r->rightNodeId = n->leftNodeId; 493 | n->leftNodeId = rn; 494 | updateNodeHeight(rn); 495 | updateNodeStat(rn); 496 | updateNodeHeight(nodeId); 497 | updateNodeStat(nodeId); 498 | } 499 | /////////////////////// 500 | 501 | 502 | bool treesort(tuple& p1, tuple& p2){ 503 | return std::get<0>(p1) == std::get<0>(p2) ? 504 | std::get<1>(p1).length() < std::get<1>(p2).length() : (std::get<0>(p2) < std::get<0>(p1)); 505 | } 506 | tuple* TreeIndex::find(wstring txt){ 507 | long nodeId = 0; 508 | Node* n = this->get(nodeId); 509 | if (n == NULL) 510 | return NULL; 511 | 512 | bool did = true; 513 | while (did && n != NULL && (n->rightNodeId > 0 || n->leftNodeId > 0) ){ 514 | did = false; 515 | if (wstring(n->text) == txt){ 516 | return new tuple(n->statCount, txt,n->value); 517 | } 518 | if (n->leftNodeId > 0 && wstring(n->text) >= txt){ 519 | n = this->get(nodeId = n->leftNodeId); 520 | did = true; 521 | } 522 | else if (n->rightNodeId > 0 && wstring(n->text) < txt){ 523 | n = this->get(nodeId = n->rightNodeId); 524 | did = true; 525 | } 526 | } 527 | return NULL; 528 | } 529 | vector > TreeIndex::search(wstring txt, int maxCount){ 530 | // std::transform(txt.begin(), txt.end(), txt.begin(), ::tolower); 531 | 532 | vector > rtn; 533 | long nodeId = 0; 534 | Node* n = this->get(nodeId); 535 | if (n == NULL) 536 | return rtn; 537 | wcout << txt << endl; 538 | bool did = true; 539 | while (did && n != NULL && (n->rightNodeId > 0 || n->leftNodeId > 0) && !hasPrefix(wstring(n->text), txt)){ 540 | did = false; 541 | if (n->leftNodeId > 0 && wstring(n->text) >= txt){ 542 | n = this->get(nodeId = n->leftNodeId); 543 | did = true; 544 | } 545 | else if (n->rightNodeId > 0 && wstring(n->text) < txt){ 546 | n = this->get(nodeId = n->rightNodeId); 547 | did = true; 548 | } 549 | } 550 | if (!hasPrefix(wstring(n->text), txt)) return rtn; 551 | 552 | priority_queue > maxq; 553 | priority_queue > curq; // has the least stat value on top maxCount 554 | 555 | maxq.push(make_pair(n->maxStatCount, nodeId)); 556 | while (!maxq.empty()){ 557 | auto t = maxq.top(); 558 | maxq.pop(); 559 | 560 | n = this->get(t.second); 561 | curq.push(make_pair(-n->statCount, t.second)); 562 | // cout << n->maxStatCount << " - " << n->statCount << " - " << -curq.top().first << endl; 563 | 564 | if (curq.size() >= maxCount && n->statCount <= -this->get(curq.top().second)->statCount){ 565 | break; 566 | } 567 | 568 | while (curq.size() > maxCount) curq.pop(); 569 | 570 | Node* nt = this->get(nodeId = n->leftNodeId); 571 | while (nt && !hasPrefix(wstring(nt->text), txt) ){ 572 | nt = this->get(nodeId = nt->rightNodeId); 573 | } 574 | if (nt && hasPrefix(wstring(nt->text), txt)) 575 | maxq.push(make_pair(nt->maxStatCount, nodeId)); 576 | 577 | nt = this->get(nodeId = n->rightNodeId); 578 | while (nt && !hasPrefix(wstring(nt->text), txt) ){ 579 | nt = this->get(nodeId = nt->leftNodeId); 580 | } 581 | if (nt && hasPrefix(wstring(nt->text), txt)) 582 | maxq.push(make_pair(nt->maxStatCount, nodeId)); 583 | 584 | 585 | 586 | 587 | } 588 | //popping the last item if exists 589 | while (curq.size() > maxCount) curq.pop(); 590 | 591 | while (!curq.empty()) 592 | { 593 | n = this->get(curq.top().second); 594 | curq.pop(); 595 | rtn.insert(rtn.begin(),tuple(n->statCount, wstring(n->text), n->value)); 596 | } 597 | //reverse(rtn.begin(), rtn.end()); 598 | 599 | //sort(rtn.begin(), rtn.end(), treesort); 600 | cout << rtn.size() << endl; 601 | return rtn; 602 | } 603 | vector TreeIndex::searchOnlyString(wstring txt, int maxCount ){ 604 | auto t=search(txt,maxCount); 605 | vector rtn; 606 | for (size_t i = 0; i < t.size(); i++) 607 | { 608 | rtn.push_back(std::get<1>(t[i])); 609 | } 610 | return rtn; 611 | } 612 | 613 | void TreeIndex::setDoubleKeys(vector& chars ){ 614 | mapDoubleKeys.clear(); 615 | for (int i=0; i > TreeIndex::searchDouble(wstring txt, int maxCount = 100){ 622 | vector > rtn; 623 | long nodeId = 0; 624 | Node* n = this->get(nodeId); 625 | if (n == NULL) 626 | return rtn; 627 | wcout << txt << endl; 628 | bool did = true; 629 | while (did && n != NULL && (n->rightNodeId > 0 || n->leftNodeId > 0) && !hasPrefix(wstring(n->text), txt)){ 630 | did = false; 631 | if (n->leftNodeId > 0 && wstring(n->text) >= txt){ 632 | n = this->get(nodeId = n->leftNodeId); 633 | did = true; 634 | } 635 | else if (n->rightNodeId > 0 && wstring(n->text) < txt){ 636 | n = this->get(nodeId = n->rightNodeId); 637 | did = true; 638 | } 639 | } 640 | if (!hasPrefix(wstring(n->text), txt)) return rtn; 641 | 642 | priority_queue > maxq; 643 | priority_queue > curq; // has the least stat value on top maxCount 644 | 645 | maxq.push(make_pair(n->maxStatCount, nodeId)); 646 | while (!maxq.empty()){ 647 | auto t = maxq.top(); 648 | maxq.pop(); 649 | 650 | n = this->get(t.second); 651 | curq.push(make_pair(-n->statCount, t.second)); 652 | // cout << n->maxStatCount << " - " << n->statCount << " - " << -curq.top().first << endl; 653 | 654 | if (curq.size() >= maxCount && n->statCount <= -this->get(curq.top().second)->statCount){ 655 | break; 656 | } 657 | 658 | while (curq.size() > maxCount) curq.pop(); 659 | 660 | Node* nt = this->get(nodeId = n->leftNodeId); 661 | while (nt && !hasPrefix(wstring(nt->text), txt) ){ 662 | nt = this->get(nodeId = nt->rightNodeId); 663 | } 664 | if (nt && hasPrefix(wstring(nt->text), txt)) 665 | maxq.push(make_pair(nt->maxStatCount, nodeId)); 666 | 667 | nt = this->get(nodeId = n->rightNodeId); 668 | while (nt && !hasPrefix(wstring(nt->text), txt) ){ 669 | nt = this->get(nodeId = nt->leftNodeId); 670 | } 671 | if (nt && hasPrefix(wstring(nt->text), txt)) 672 | maxq.push(make_pair(nt->maxStatCount, nodeId)); 673 | 674 | 675 | 676 | 677 | } 678 | //popping the last item if exists 679 | while (curq.size() > maxCount) curq.pop(); 680 | 681 | while (!curq.empty()) 682 | { 683 | n = this->get(curq.top().second); 684 | curq.pop(); 685 | rtn.insert(rtn.begin(),tuple(n->statCount, wstring(n->text), n->value)); 686 | } 687 | //reverse(rtn.begin(), rtn.end()); 688 | 689 | //sort(rtn.begin(), rtn.end(), treesort); 690 | cout << rtn.size() << endl; 691 | return rtn; 692 | 693 | } 694 | } -------------------------------------------------------------------------------- /TreeIndex.h: -------------------------------------------------------------------------------- 1 | #include 2 | #include 3 | #include 4 | #include 5 | #include 6 | using namespace std; 7 | 8 | namespace Index{ 9 | class Node{ 10 | public: 11 | Node(); 12 | //~Node(); 13 | wchar_t* text; 14 | long textIndex; 15 | int textCount; 16 | long leftNodeId; 17 | long rightNodeId; 18 | int nodeHeight; 19 | long statCount; 20 | long maxStatCount; 21 | long value; 22 | int valueCount; 23 | long parentTextNodeId; 24 | 25 | 26 | }/*__attribute__((packed))*/; 27 | 28 | class TreeIndex{ 29 | protected: 30 | int balanceFactor(long nodeId); 31 | int nodeHeight(long nodeId); 32 | int updateNodeHeight(long nodeId); 33 | int updateNodeStat(long nodeId); 34 | bool balanceNode(long nodeId); 35 | void rotateRight(long nodeId); 36 | void rotateLeft(long nodeId); 37 | /////////////////////// 38 | Node* get(long nodeId); 39 | long newNode(); 40 | FILE* file; 41 | FILE* wordFile; 42 | map cache; 43 | set saveQueue; 44 | //TreeIndex(); 45 | long _size; 46 | 47 | void updateMaxStatCounts(vector& nodes); 48 | string fileName; 49 | ///double keys 50 | map mapDoubleKeys; 51 | public: 52 | TreeIndex(const string fileName); 53 | TreeIndex(); 54 | ~TreeIndex(); 55 | void save(); 56 | long size(); 57 | vector > search(wstring txt, int maxCount = 100); 58 | vector searchOnlyString(wstring txt, int maxCount = 100); 59 | tuple* find(wstring txt); 60 | void add(wstring txt, long statCount = 1/*rank*/,long value = 0); 61 | void addSave(wstring txt, long statCount = 1); 62 | // void incrementUsage(wstring txt, int statCount = 1); 63 | void loadNgram(string fileName, int ngram); 64 | void loadTopWords(string fileName,int maxCount=-1); 65 | bool isExist(wstring s); 66 | void setText(Node* n, wstring txt); 67 | void clearCache(); 68 | int cacheSize(); 69 | int treeHeight(); 70 | bool ensureStats(int node=0,long stat=0); 71 | 72 | /// double keys 73 | void setDoubleKeys(vector& chars ); 74 | vector > searchDouble(wstring txt, int maxCount = 100); 75 | 76 | }; 77 | 78 | class MultiNode{ 79 | public: 80 | MultiNode(); 81 | wchar_t* text; 82 | long textIndex; 83 | int textCount; 84 | long statCount; 85 | }; 86 | 87 | class MultiTreeIndex:TreeIndex{ 88 | protected: 89 | FILE* multiNodeFile; 90 | public: 91 | //MultiValueTreeIndex(wstring fileName); 92 | //MultiValueTreeIndex(); 93 | //~MultiValueTreeIndex(); 94 | void add(wstring txt,wstring value, long statCount = 1); 95 | vector > search(wstring txt, int maxCount = 100); 96 | vector searchOnlyString(wstring txt, int maxCount = 100); 97 | vector > find(wstring txt); 98 | 99 | }; 100 | 101 | } -------------------------------------------------------------------------------- /main.cpp: -------------------------------------------------------------------------------- 1 | #include 2 | #include 3 | #include 4 | #include 5 | #include "TreeIndex.h" 6 | using namespace std; 7 | using namespace Index; 8 | 9 | int main(){ 10 | 11 | 12 | 13 | 14 | const int maxCount = 1000000; 15 | // TreeIndex en("index_en1.tf"); 16 | // en.loadTopWords("./langs/en.txt", maxCount); 17 | // return 0; 18 | /* 19 | en.loadNgram(".\\ngrams\\t.txt", 2); 20 | 21 | TreeIndex ru(L"index_ru.tf"); 22 | ru.loadTopWords(".\\langs\\ru.txt", maxCount); 23 | 24 | 25 | 26 | TreeIndex fr(L"index_fr.tf"); 27 | fr.loadTopWords(".\\langs\\fr.txt", maxCount); 28 | 29 | TreeIndex de(L"index_de.tf"); 30 | de.loadTopWords(".\\langs\\de.txt", maxCount); 31 | 32 | TreeIndex es(L"index_es.tf"); 33 | es.loadTopWords(".\\langs\\es.txt", maxCount); 34 | 35 | TreeIndex tr(L"index_tr.tf"); 36 | tr.loadTopWords(".\\langs\\tr.txt", maxCount); 37 | 38 | TreeIndex da(L"index_da.tf"); 39 | da.loadTopWords(".\\langs\\da.txt", maxCount); 40 | 41 | TreeIndex pl(L"index_pl.tf"); 42 | pl.loadTopWords(".\\langs\\pl.txt", maxCount); 43 | 44 | TreeIndex pt(L"index_pt.tf"); 45 | pt.loadTopWords(".\\langs\\pt.txt", maxCount); 46 | 47 | 48 | 49 | TreeIndex sv(L"index_sv.tf"); 50 | sv.loadTopWords(".\\langs\\sv.txt", maxCount); 51 | 52 | 53 | return 0;*/ 54 | 55 | 56 | TreeIndex xx("index_en1.tf"); 57 | cout << xx.treeHeight() << endl; 58 | 59 | chrono::time_point start,end; 60 | start = chrono::system_clock::now(); 61 | 62 | // wifstream in("./langs/en.txt"); 63 | // wstring word; 64 | // int counter = 1; 65 | // //in.imbue(std::locale(std::locale::classic(), new std::codecvt_utf8)); 66 | // 67 | // //maxCount = 100000; 68 | // 69 | // wstring prefix=L"air"; 70 | // vector lst; 71 | // while (in >> word && counter<=maxCount){ 72 | // lst.push_back(word); 73 | // if(word.size()>=prefix.size() && word.compare(0,prefix.size(),prefix)==0){ 74 | // wcout<=prefix.size() && word.compare(0,prefix.size(),prefix)==0){ 83 | // wcout<(xff[j]) << endl; 94 | } 95 | if (xff.size()>0){ 96 | wcout<(xff[0]) << endl; 97 | } 98 | 99 | } 100 | 101 | end = chrono::system_clock::now(); 102 | chrono::duration d = end - start; 103 | cout<