Welcome to mirror list, hosted at ThFree Co, Russian Federation.

github.com/marian-nmt/sentencepiece.git - Unnamed repository; edit this file 'description' to name the repository.
summaryrefslogtreecommitdiff
path: root/src
diff options
context:
space:
mode:
authorTaku Kudo <taku910@users.noreply.github.com>2018-05-03 18:11:17 +0300
committerGitHub <noreply@github.com>2018-05-03 18:11:17 +0300
commit77ec954d23591cd5180f4d6160c7862c600c4b09 (patch)
tree5d01d0f3b80428a3c1b2ae58594c5935e6f90d0d /src
parent63983329c59a52366ce39031065e84442e32f66b (diff)
Update sentencepiece_processor.h
Diffstat (limited to 'src')
-rw-r--r--src/sentencepiece_processor.h6
1 files changed, 4 insertions, 2 deletions
diff --git a/src/sentencepiece_processor.h b/src/sentencepiece_processor.h
index ff41143..30854eb 100644
--- a/src/sentencepiece_processor.h
+++ b/src/sentencepiece_processor.h
@@ -197,8 +197,10 @@ class SentencePieceProcessor {
// forward-filtering and backward-sampling algorithm.
// `alpha` is a smoothing parameter. The best segmentation
// (Viterbi segmentation) is more likely sampled when setting larger
- // alpha (alpha >= 1.0). When alpha is 0.0, one segmentation is
- // uniformly sampled from the nbest or lattice.
+ // alpha. When alpha is 0.0, one segmentation is uniformly sampled from the
+ // nbest or lattice.
+ // `nbest_size` and `alpha` correspond to parameters `l` and `alpha`
+ // in https://arxiv.org/abs/1804.10959 (nbest_size < 0 means l = infinity)
virtual util::Status SampleEncode(const std::string &input, int nbest_size,
float alpha,
std::vector<std::string> *pieces) const;