mosesdecoder/mert/BleuScorer.cpp

#include "BleuScorer.h"

#include <algorithm>
#include <cmath>
#include <climits>
#include <fstream>
#include <iterator>
#include <stdexcept>
#include "Util.h"

BleuScorer::BleuScorer(const string& config)
    : StatisticsBasedScorer("BLEU",config),
      kLENGTH(4),
      m_ref_length_type(CLOSEST) {
  //configure regularisation
  static string KEY_REFLEN = "reflen";
  static string REFLEN_AVERAGE = "average";
  static string REFLEN_SHORTEST = "shortest";
  static string REFLEN_CLOSEST = "closest";

  string reflen = getConfig(KEY_REFLEN,REFLEN_CLOSEST);
  if (reflen == REFLEN_AVERAGE) {
    m_ref_length_type = AVERAGE;
  } else if (reflen == REFLEN_SHORTEST) {
    m_ref_length_type = SHORTEST;
  } else if (reflen == REFLEN_CLOSEST) {
    m_ref_length_type = CLOSEST;
  } else {
    throw runtime_error("Unknown reference length strategy: " + reflen);
  }
  //    cerr << "Using reference length strategy: " << reflen << endl;
}

BleuScorer::~BleuScorer() {}

size_t BleuScorer::countNgrams(const string& line, counts_t& counts, unsigned int n)
{
  vector<int> encoded_tokens;
  //cerr << line << endl;
  TokenizeAndEncode(line, encoded_tokens);
  //copy(encoded_tokens.begin(), encoded_tokens.end(), ostream_iterator<int>(cerr," "));
  //cerr << endl;
  for (size_t k = 1; k <= n; ++k) {
    //ngram order longer than sentence - no point
    if (k > encoded_tokens.size()) {
      continue;
    }
    for (size_t i = 0; i < encoded_tokens.size()-k+1; ++i) {
      vector<int> ngram;
      for (size_t j = i; j < i+k && j < encoded_tokens.size(); ++j) {
        ngram.push_back(encoded_tokens[j]);
      }
      int count = 1;
      counts_iterator oldcount = counts.find(ngram);
      if (oldcount != counts.end()) {
        count = (oldcount->second) + 1;
      }
      //cerr << count << endl;
      counts[ngram] = count;
      //cerr << endl;
    }
  }
  //cerr << "counted ngrams" << endl;
  //dump_counts(counts);
  return encoded_tokens.size();
}

void BleuScorer::setReferenceFiles(const vector<string>& referenceFiles)
{
  //make sure reference data is clear
  m_ref_counts.reset();
  m_ref_lengths.clear();
  ClearEncoder();

  //load reference data
  for (size_t i = 0; i < referenceFiles.size(); ++i) {
    TRACE_ERR("Loading reference from " << referenceFiles[i] << endl);
    ifstream refin(referenceFiles[i].c_str());
    if (!refin) {
      throw runtime_error("Unable to open: " + referenceFiles[i]);
    }
    string line;
    size_t sid = 0; //sentence counter
    while (getline(refin,line)) {
      //cerr << line << endl;
      if (i == 0) {
        counts_t *counts = new counts_t; //these get leaked
        m_ref_counts.push_back(counts);
        vector<size_t> lengths;
        m_ref_lengths.push_back(lengths);
      }
      if (m_ref_counts.size() <= sid) {
        throw runtime_error("File " + referenceFiles[i] + " has too many sentences");
      }
      counts_t counts;
      size_t length = countNgrams(line,counts,kLENGTH);
      //for any counts larger than those already there, merge them in
      for (counts_iterator ci = counts.begin(); ci != counts.end(); ++ci) {
        counts_iterator oldcount_it = m_ref_counts[sid]->find(ci->first);
        int oldcount = 0;
        if (oldcount_it != m_ref_counts[sid]->end()) {
          oldcount = oldcount_it->second;
        }
        int newcount = ci->second;
        if (newcount > oldcount) {
          m_ref_counts[sid]->operator[](ci->first) = newcount;
        }
      }
      //add in the length
      m_ref_lengths[sid].push_back(length);
      if (sid > 0 && sid % 100 == 0) {
        TRACE_ERR(".");
      }
      ++sid;
    }
    TRACE_ERR(endl);
  }
}


void BleuScorer::prepareStats(size_t sid, const string& text, ScoreStats& entry)
{
//      cerr << text << endl;
//      cerr << sid << endl;
  //dump_counts(*m_ref_counts[sid]);
  if (sid >= m_ref_counts.size()) {
    stringstream msg;
    msg << "Sentence id (" << sid << ") not found in reference set";
    throw runtime_error(msg.str());
  }
  counts_t testcounts;
  //stats for this line
  vector<float> stats(kLENGTH*2);;
  size_t length = countNgrams(text,testcounts,kLENGTH);
  //dump_counts(testcounts);
  if (m_ref_length_type == SHORTEST) {
    //cerr << reflengths.size() << " " << sid << endl;
    int shortest = *min_element(m_ref_lengths[sid].begin(), m_ref_lengths[sid].end());
    stats.push_back(shortest);
  } else if (m_ref_length_type == AVERAGE) {
    int total = 0;
    for (size_t i = 0; i < m_ref_lengths[sid].size(); ++i) {
      total += m_ref_lengths[sid][i];
    }
    const float mean = static_cast<float>(total) / m_ref_lengths[sid].size();
    stats.push_back(mean);
  } else if (m_ref_length_type == CLOSEST)  {
    int min_diff = INT_MAX;
    int min_idx = 0;
    for (size_t i = 0; i < m_ref_lengths[sid].size(); ++i) {
      const int reflength = m_ref_lengths[sid][i];
      const int diff = reflength - static_cast<int>(length);
      const int absolute_diff = abs(diff) - abs(min_diff);

      if (absolute_diff < 0) { //look for the closest reference
        min_diff = diff;
        min_idx = i;
      } else if (absolute_diff == 0) { // if two references has the same closest length, take the shortest
        if (reflength < static_cast<int>(m_ref_lengths[sid][min_idx])) {
          min_idx = i;
        }
      }
    }
    stats.push_back(m_ref_lengths[sid][min_idx]);
  } else {
    throw runtime_error("Unsupported reflength strategy");
  }
  //cerr << "computed length" << endl;
  //precision on each ngram type
  for (counts_iterator testcounts_it = testcounts.begin();
       testcounts_it != testcounts.end(); ++testcounts_it) {
    counts_iterator refcounts_it = m_ref_counts[sid]->find(testcounts_it->first);
    int correct = 0;
    int guess = testcounts_it->second;
    if (refcounts_it != m_ref_counts[sid]->end()) {
      correct = min(refcounts_it->second,guess);
    }
    size_t len = testcounts_it->first.size();
    stats[len*2-2] += correct;
    stats[len*2-1] += guess;
  }
  stringstream sout;
  copy(stats.begin(),stats.end(),ostream_iterator<float>(sout," "));
  //TRACE_ERR(sout.str() << endl);
  string stats_str = sout.str();
  entry.set(stats_str);
}

float BleuScorer::calculateScore(const vector<int>& comps) const
{
  //cerr << "BLEU: ";
  //copy(comps.begin(),comps.end(), ostream_iterator<int>(cerr," "));
  float logbleu = 0.0;
  for (int i = 0; i < kLENGTH; ++i) {
    if (comps[2*i] == 0) {
      return 0.0;
    }
    logbleu += log(comps[2*i]) - log(comps[2*i+1]);

  }
  logbleu /= kLENGTH;
  const float brevity = 1.0 - static_cast<float>(comps[kLENGTH*2]) / comps[1];//reflength divided by test length
  if (brevity < 0.0) {
    logbleu += brevity;
  }
  //cerr << " " << exp(logbleu) << endl;
  return exp(logbleu);
}

void BleuScorer::dump_counts(counts_t& counts) const {
  for (counts_const_iterator i = counts.begin(); i != counts.end(); ++i) {
    cerr << "(";
    copy(i->first.begin(), i->first.end(), ostream_iterator<int>(cerr," "));
    cerr << ") " << i->second << ", ";
  }
  cerr << endl;
}