diff options
Diffstat (limited to 'klm/search/rule.cc')
-rw-r--r-- | klm/search/rule.cc | 43 |
1 files changed, 43 insertions, 0 deletions
diff --git a/klm/search/rule.cc b/klm/search/rule.cc new file mode 100644 index 00000000..5b00207e --- /dev/null +++ b/klm/search/rule.cc @@ -0,0 +1,43 @@ +#include "search/rule.hh" + +#include "search/context.hh" +#include "search/final.hh" + +#include <ostream> + +#include <math.h> + +namespace search { + +template <class Model> float ScoreRule(const Context<Model> &context, const std::vector<lm::WordIndex> &words, bool prepend_bos, lm::ngram::ChartState *writing) { + unsigned int oov_count = 0; + float prob = 0.0; + const Model &model = context.LanguageModel(); + const lm::WordIndex oov = model.GetVocabulary().NotFound(); + for (std::vector<lm::WordIndex>::const_iterator word = words.begin(); ; ++word) { + lm::ngram::RuleScore<Model> scorer(model, *(writing++)); + // TODO: optimize + if (prepend_bos && (word == words.begin())) { + scorer.BeginSentence(); + } + for (; ; ++word) { + if (word == words.end()) { + prob += scorer.Finish(); + return static_cast<float>(oov_count) * context.GetWeights().OOV() + prob * context.GetWeights().LM(); + } + if (*word == kNonTerminal) break; + if (*word == oov) ++oov_count; + scorer.Terminal(*word); + } + prob += scorer.Finish(); + } +} + +template float ScoreRule(const Context<lm::ngram::RestProbingModel> &model, const std::vector<lm::WordIndex> &words, bool prepend_bos, lm::ngram::ChartState *writing); +template float ScoreRule(const Context<lm::ngram::ProbingModel> &model, const std::vector<lm::WordIndex> &words, bool prepend_bos, lm::ngram::ChartState *writing); +template float ScoreRule(const Context<lm::ngram::TrieModel> &model, const std::vector<lm::WordIndex> &words, bool prepend_bos, lm::ngram::ChartState *writing); +template float ScoreRule(const Context<lm::ngram::QuantTrieModel> &model, const std::vector<lm::WordIndex> &words, bool prepend_bos, lm::ngram::ChartState *writing); +template float ScoreRule(const Context<lm::ngram::ArrayTrieModel> &model, const std::vector<lm::WordIndex> &words, bool prepend_bos, lm::ngram::ChartState *writing); +template float ScoreRule(const Context<lm::ngram::QuantArrayTrieModel> &model, const std::vector<lm::WordIndex> &words, bool prepend_bos, lm::ngram::ChartState *writing); + +} // namespace search |