From d3e2ec203a5cf550320caa8023ac3dd103b0be7d Mon Sep 17 00:00:00 2001 From: Chris Dyer Date: Mon, 13 Oct 2014 00:42:37 -0400 Subject: new kenlm --- klm/lm/neural/wordvecs.cc | 23 +++++++++++++++++++++++ 1 file changed, 23 insertions(+) create mode 100644 klm/lm/neural/wordvecs.cc (limited to 'klm/lm/neural/wordvecs.cc') diff --git a/klm/lm/neural/wordvecs.cc b/klm/lm/neural/wordvecs.cc new file mode 100644 index 00000000..09bb4260 --- /dev/null +++ b/klm/lm/neural/wordvecs.cc @@ -0,0 +1,23 @@ +#include "lm/neural/wordvecs.hh" + +#include "util/file_piece.hh" + +namespace lm { namespace neural { + +WordVecs::WordVecs(util::FilePiece &f) { + const unsigned long lines = f.ReadULong(); + const std::size_t vocab_mem = ngram::ProbingVocabulary::Size(lines, 1.5); + vocab_backing_.reset(util::CallocOrThrow(vocab_mem)); + vocab_.SetupMemory(vocab_backing_.get(), vocab_mem); + const unsigned long width = f.ReadULong(); + vecs_.resize(width, lines); + for (unsigned long i = 0; i < lines; ++i) { + WordIndex column = vocab_.Insert(f.ReadDelimited()); + for (unsigned int row = 0; row < width; ++row) { + vecs_(row,column) = f.ReadFloat(); + } + } + vocab_.FinishedLoading(); +} + +}} // namespaces -- cgit v1.2.3