Rewrote quite a bit of kgramstats

The algorithm still treats most tokens literally, but now groups together tokens that terminate a clause somehow (so, contain .?!,), without distinguishing between the different terminating characters. For each word that can terminate a sentence, the algorithm creates a histogram of the terminating characters and number of occurrences of those characters for that word (number of occurrences is to allow things like um???? and um,,,,, to still be folded down into um.). Then, when the terminating version of that token is invoked, a random terminating string is added to that token based on the histogram for that word (again, to allow things like the desu-ly use of multiple commas to end clauses).

The algorithm now also has a slightly advanced kgram structure; a special "sentence wildcard" kgram value is set aside from normal strings of tokens that can match any terminating token. This kgram value is never printed (it is only ever present in the query kgrams and cannot actually be present in the histograms (it is of a different datatype)) and is used at the beginning of sentence generation to make sure that the first couple of words generated actually form the beginning of a sentence instead of picking up somewhere in the middle of a sentence. It is also used to reset sentence generation in the rare occasion that the end of the corpus is reached.

1 files changed, 165 insertions, 277 deletions

@@ -5,237 +5,176 @@
 #include <algorithm>
 #include "malaprop.h"
 
+query wildcardQuery(querytype_sentence);
+
 std::string canonize(std::string f);
 
 // runs in O(t^2) time where t is the number of tokens in the input corpus
 // We consider maxK to be fairly constant
 kgramstats::kgramstats(std::string corpus, int maxK)
 {
-        this->maxK = maxK;
+  this->maxK = maxK;
   
   std::vector<std::string> tokens;
-    size_t start = 0;
-        int end = 0;
+  size_t start = 0;
+  int end = 0;
 
-        while (end != std::string::npos)
-        {
-           end = corpus.find(" ", start);
+  while (end != std::string::npos)
+  {
+    end = corpus.find(" ", start);
 
-       std::string token = corpus.substr(start, (end == std::string::npos) ? std::string::npos : end - start);
-       if (token[token.length()-1] == '\n')
-       {
-         if ((token[token.length()-2] != '.') && (token[token.length()-2] != '!') && (token[token.length()-2] != '?'))
-         {
-           token.insert(token.length()-1, ".");
-         }
+    std::string token = corpus.substr(start, (end == std::string::npos) ? std::string::npos : end - start);
+    if (token[token.length()-1] == '\n')
+    {
+      if ((token[token.length()-2] != '.') && (token[token.length()-2] != '!') && (token[token.length()-2] != '?') && (token[token.length()-2] != ','))
+      {
+        token.insert(token.length()-1, ".");
+      }
          
-         token.resize(token.length()-1);
-       }
+      token.resize(token.length()-1);
+    }
        
-       if (token.compare("") && token.compare("."))
-       {
-         mstats.addWord(token);
-           tokens.push_back(token);
-       }
+    if (token.compare("") && token.compare("."))
+    {
+      mstats.addWord(token);
+      tokens.push_back(token);
+    }
 
-           start = ((end > (std::string::npos - 1) ) ? std::string::npos : end + 1);
-        }
+    start = ((end > (std::string::npos - 1) ) ? std::string::npos : end + 1);
+  }
         
-        std::map<kgram, std::map<std::string, token_data*>* > tstats;
-  bool newSentence = true;
-  bool newClause = false;
-        for (int k=0; k<maxK; k++)
-        {
-                for (int i=0; i<(tokens.size() - k); i++)
-                {
-                        kgram seq(tokens.begin()+i, tokens.begin()+i+k);
-                        std::transform(seq.begin(), seq.end(), seq.begin(), canonize);
-                        std::string f = tokens[i+k];
-                        
-      
-      
-      std::string canonical = canonize(f);
-                        
-                        if (tstats[seq] == NULL)
-                        {
-                                tstats[seq] = new std::map<std::string, token_data*>();
-                        }
-                        
-                        if ((*tstats[seq])[canonical] == NULL)
-                        {
-                                (*tstats[seq])[canonical] = (token_data*) calloc(1, sizeof(token_data));
-                        }
-
-                        token_data* td = tstats[seq]->at(canonical);
-                        td->token = new std::string(canonical);
-                        td->all++;
+  std::map<kgram, std::map<token, token_data> > tstats;
+  std::map<token, std::map<termstats, int> > tendings;
+  for (int k=1; k<maxK; k++)
+  {
+    for (int i=0; i<(tokens.size() - k); i++)
+    {
+      std::list<std::string> seq(tokens.begin()+i, tokens.begin()+i+k);
+      kgram prefix;
       
-      /*if (newSentence)
+      for (std::list<std::string>::iterator it = seq.begin(); it != seq.end(); it++)
       {
-        kgram newKgram(1, ".");
-        if (tstats[newKgram] == NULL)
+        token word(canonize(*it));
+        
+        if (it->find_first_of(".?!,") != std::string::npos)
         {
-          tstats[newKgram] = new std::map<std::string, token_data*>();
+          word.terminating = true;
         }
         
-        (*tstats[newKgram])[canonical] = td;
-        
-        newSentence = false;
+        prefix.push_back(word);
       }
       
-      if (newClause)
+      std::string f = tokens[i+k];
+                  std::string canonical = canonize(f);
+      
+      token word(canonical);
+      if (f.find_first_of(".?!,") != std::string::npos)
       {
-        kgram commaKgram(1, ",");
-        if (tstats[commaKgram] == NULL)
-        {
-          tstats[commaKgram] = new std::map<std::string, token_data*>();
-        }
+        word.terminating = true;
         
-        (*tstats[commaKgram])[canonical] = td;
+        char terminator = f[f.find_last_of(".?!,")];
+        int occurrences = std::count(f.begin(), f.end(), terminator);
         
-        newClause = false;
-      }
-      
-      if ((f.length() > 0) && (f[f.length()-1] == '\n'))
-      {
-        td->period++;
-        newSentence = true;
-        f.resize(f.length()-1);
+        tendings[word][termstats(terminator, occurrences)]++;
       }
                         
-      if (f.length() > 0)
+      token_data& td = tstats[prefix][word];
+      td.word = word;
+      td.all++;
+
+      if (std::find_if(f.begin(), f.end(), ::islower) == f.end())
       {
-                        if ((f[f.length()-1] == '.') || (f[f.length()-1] == '!') || (f[f.length()-1] == '?'))
-                        {
-          if (!newSentence)
-          {
-                                td->period++;
-            newSentence = true;
-          }
-                                
-          f.resize(f.length()-1);
-                        } else if (f[f.length()-1] == ',')
-        {
-          if (!newSentence)
-          {
-            td->comma++;
-            newClause = true;
-          }
-          
-          f.resize(f.length()-1);
-        }
+        td.uppercase++;
+      } else if (isupper(f[0]))
+      {
+        td.titlecase++;
       }
       
-      if (f.length() > 0)
+      if (prefix.front().word.terminating)
       {
-        if (f[0] == '"')
-        {
-          td->startquote++;
-        }
+        prefix.front() = wildcardQuery;
         
-        if (f[0] == '(')
+        token_data& td2 = tstats[prefix][word];
+        td2.word = word;
+        td2.all++;
+
+        if (std::find_if(f.begin(), f.end(), ::islower) == f.end())
         {
-          td->startparen++;
-        }
-        
-        if ((f[f.length()-1] == '"') || (f[f.length()-1] == ')'))
+          td2.uppercase++;
+        } else if (isupper(f[0]))
         {
-          if (f[f.length()-1] == '"')
-          {
-            td->endquote++;
-          } else if (f[f.length()-1] == ')')
-          {
-            td->endparen++;
-          }
-          
-          f.resize(f.length()-1);
-          
-          if (f.length() > 0)
-          {
-                        if ((f[f.length()-1] == '.') || (f[f.length()-1] == '!') || (f[f.length()-1] == '?'))
-                        {
-              if (!newSentence)
-              {
-                                        td->period++;
-                newSentence = true;
-              }
-                        } else if (f[f.length()-1] == ',')
-            {
-              if (!newSentence && !newClause)
-              {
-                td->comma++;
-                newClause = true;
-              }
-            }
-          }
-        }
-      }*/
-                        
-                        if (std::find_if(f.begin(), f.end(), ::islower) == f.end())
-                        {
-                                td->uppercase++;
-                        } else if (isupper(f[0]))
-                        {
-                                td->titlecase++;
-                        }
-      
-      /*if (k != 0)
-      {
-        if (newSentence)
-        {
-          i += k;
+          td2.titlecase++;
         }
-        
-        newSentence = false;
-        newClause = false;
-      }*/
-                }
-        }
+      }
+    }
+  }
         
-        stats = new std::map<kgram, std::map<int, token_data*>* >();
-        for (std::map<kgram, std::map<std::string, token_data*>* >::iterator it = tstats.begin(); it != tstats.end(); it++)
-        {
-                kgram klist = it->first;
-                std::map<std::string, token_data*>* probtable = it->second;
-                std::map<int, token_data*>* distribution = new std::map<int, token_data*>();
-        int max = 0;
+  for (std::map<kgram, std::map<token, token_data> >::iterator it = tstats.begin(); it != tstats.end(); it++)
+  {
+    kgram klist = it->first;
+    std::map<token, token_data>& probtable = it->second;
+    std::map<int, token_data>& distribution = stats[klist];
+    int max = 0;
                 
-                for (std::map<std::string, token_data*>::iterator kt = probtable->begin(); kt != probtable->end(); kt++)
-                {
-                        max += kt->second->all;
+    for (std::map<token, token_data>::iterator kt = probtable.begin(); kt != probtable.end(); kt++)
+    {
+      max += kt->second.all;
                         
-                        (*distribution)[max] = kt->second;
-                }
-                
-                (*stats)[klist] = distribution;
-        }
+      distribution[max] = kt->second;
+    }
+  }
+  
+  for (std::map<token, std::map<termstats, int> >::iterator it = tendings.begin(); it != tendings.end(); it++)
+  {
+    token word = it->first;
+    std::map<termstats, int>& probtable = it->second;
+    std::map<int, termstats>& distribution = endings[word];
+    int max = 0;
+    
+    for (std::map<termstats, int>::iterator kt = probtable.begin(); kt != probtable.end(); kt++)
+    {
+      max += kt->second;
+      
+      distribution[max] = kt->first;
+    }
+  }
 }
 
 void printKgram(kgram k)
 {
-        for (kgram::iterator it = k.begin(); it != k.end(); it++)
-        {
-                std::cout << *it << " ";
-        }
+  for (kgram::iterator it = k.begin(); it != k.end(); it++)
+  {
+    query& q = *it;
+    if (q.type == querytype_sentence)
+    {
+      std::cout << "#.# ";
+    } else if (q.type == querytype_literal)
+    {
+      if (q.word.terminating)
+      {
+        std::cout << q.word.canon << ". ";
+      } else {
+        std::cout << q.word.canon << " ";
+      }
+    }
+  }
 }
 
 // runs in O(n log t) time where n is the input number of sentences and t is the number of tokens in the input corpus
 std::vector<std::string> kgramstats::randomSentence(int n)
 {
-        std::vector<std::string> result;
-  kgram newKgram(1, ".");
-  kgram commaKgram(1, ",");
-        std::list<std::string> cur;
+  std::vector<std::string> result;
+  kgram cur(1, wildcardQuery);
   int cuts = 0;
         
-        for (int i=0; i<n; i++)
-        {
-                if (cur.size() == maxK)
-                {
-                        cur.pop_front();
-                }
+  for (int i=0; i<n; i++)
+  {
+    if (cur.size() == maxK)
+    {
+      cur.pop_front();
+    }
     
-    if ((cur.size() > 0) && (cur != newKgram))
+    if (cur.size() > 0)
     {
       if (rand() % (maxK - cur.size() + 1) == 0)
       {
@@ -253,20 +192,19 @@ std::vector<std::string> kgramstats::randomSentence(int n)
       
       cuts++;
     }
+    
+    // Gotta circumvent the last line of the input corpus
+    // https://twitter.com/starla4444/status/684222271339237376
+    if (stats.count(cur) == 0)
+    {
+      cur = kgram(1, wildcardQuery);
+    }
 
-                std::map<int, token_data*> distribution = *(*stats)[cur];
-                int max = distribution.rbegin()->first;
-                int r = rand() % max;
-                token_data* next = distribution.upper_bound(r)->second;
-
-                std::string nextToken(*(next->token));
-                int casing = rand() % next->all;
-                /*int period = rand() % next->all;
-    int startparen = rand() % next->all;
-    int endparen = rand() % next->all;
-    int startquote = rand() % next->all;
-    int endquote = rand() % next->all;
-    int comma = rand() % next->all;*/
+    std::map<int, token_data>& distribution = stats[cur];
+    int max = distribution.rbegin()->first;
+    int r = rand() % max;
+    token_data& next = distribution.upper_bound(r)->second;
+    std::string nextToken(next.word.canon);
     
     bool mess = (rand() % 100) == 0;
     if (mess)
@@ -274,114 +212,64 @@ std::vector<std::string> kgramstats::randomSentence(int n)
       nextToken = mstats.alternate(nextToken);
     }
     
-                if (casing < next->uppercase)
-                {
-                        std::transform(nextToken.begin(), nextToken.end(), nextToken.begin(), ::toupper);
-                }
-    
-    if ((cur == newKgram) && (rand() % 15 > 0))
+    // Determine the casing of the next token. We randomly make the token all
+    // caps based on the markov chain. Otherwise, we check if the previous
+    // token is the end of a sentence (terminating token or a wildcard query).
+    int casing = rand() % next.all;
+    if (casing < next.uppercase)
+    {
+      std::transform(nextToken.begin(), nextToken.end(), nextToken.begin(), ::toupper);
+    } else if ((((cur.rbegin()->type == querytype_sentence)
+          || ((cur.rbegin()->type == querytype_literal)
+            && (cur.rbegin()->word.terminating)))
+        && (rand() % 2 > 0))
+      || (casing - next.uppercase < next.titlecase))
     {
       nextToken[0] = toupper(nextToken[0]);
     }
 
-    /*if (startquote < next->startquote)
-    {
-      nextToken = "\"" + nextToken;
-    } else if (startparen < next->startparen)
+    if (next.word.terminating)
     {
-      nextToken = "(" + nextToken;
+      std::map<int, termstats>& ending = endings[next.word];
+      int emax = ending.rbegin()->first;
+      int er = rand() % emax;
+      termstats& nextend = ending.upper_bound(er)->second;
+      
+      nextToken.append(std::string(nextend.occurrences, nextend.terminator));
     }
-        
-                if (period < next->period)
-                {
-      if (endquote < next->endquote)
-      {
-        nextToken += "\"";
-      } else if (endparen < next->endparen)
-      {
-        nextToken += ")";
-      }
-    
-      int type = rand() % 6;
-    
-      if (type < 3)
-      {
-        nextToken += ".";
-      } else if (type < 5)
-      {
-        nextToken += "!";
-      } else {
-        nextToken += "?";
-      }
-                } else if (comma < next->comma)
-    {
-      if (endquote < next->endquote)
-      {
-        nextToken += "\"";
-      } else if (endparen < next->endparen)
-      {
-        nextToken += ")";
-      }
-    
-      nextToken += ",";
-    }*/
                 
-                /* DEBUG */
-                for (kgram::iterator it = cur.begin(); it != cur.end(); it++)
-                {
-                        std::cout << *it << " ";
-                }
+    /* DEBUG */
+    printKgram(cur);
                 
-                std::cout << "-> \"" << nextToken << "\" (" << next->all << "/" << max << ")";
+    std::cout << "-> \"" << nextToken << "\" (" << next.all << "/" << max << ")";
     
     if (mess)
     {
-      std::cout << " mala " << *(next->token);
+      std::cout << " mala " << next.word.canon;
     }
     
     std::cout << std::endl;
-    
-    /*if ((cur == newKgram) || (cur == commaKgram))
-    {
-      cur.pop_front();
-    }
-                
-    if (period < next->period)// && ((rand() % 3) != 0))
-    {
-      cur = newKgram;
-    } else if ((comma < next->comma) && ((rand() % 3) == 0))
-    {
-      cur = commaKgram;
-    } else {*/
-      //if (mess && (rand() % 2 == 0))
-      if (false)
-      {
-        // This doesn't work because sometimes the alternate token isn't actually present in the original corpus
-        cur.clear();
-        cur.push_back(nextToken);
-      } else {
-        cur.push_back(*(next->token));
-      }
-      //}
+
+    cur.push_back(next.word);
                 
-                result.push_back(nextToken);
-        }
+    result.push_back(nextToken);
+  }
         
-        return result;
+  return result;
 }
 
 bool removeIf(char c)
 {
-  return !((c != '.') && (c != '?') && (c != '!') && (c != '"') && (c != '(') && (c != ')') && (c != ',') && (c != '\n'));
+  return !((c != '.') && (c != '?') && (c != '!') && (c != ',') /*&& (c != '"') && (c != '(') && (c != ')') && (c != '\n')*/);
 }
 
 std::string canonize(std::string f)
 {
-        std::string canonical(f);
-        std::transform(canonical.begin(), canonical.end(), canonical.begin(), ::tolower);
+  std::string canonical(f);
+  std::transform(canonical.begin(), canonical.end(), canonical.begin(), ::tolower);
   
   std::string result;
   std::remove_copy_if(canonical.begin(), canonical.end(), std::back_inserter(result), removeIf);
         
-        return canonical;
+  return result;
 }