diff options
| author | graehl <graehl@ec762483-ff6d-05da-a07a-a48fb63a330f> | 2010-07-23 23:10:34 +0000 |
|---|---|---|
| committer | graehl <graehl@ec762483-ff6d-05da-a07a-a48fb63a330f> | 2010-07-23 23:10:34 +0000 |
| commit | 5f9c85a6072d64b10879feafc040374e274027eb (patch) | |
| tree | cebeb55f05e770bd8b1983821312c1856caec0a8 /decoder/ff.cc | |
| parent | a4187c028432c9f9ec3693747abc1c52ef310deb (diff) | |
strip debug param from feature spec; debug info on from_fsa Init
git-svn-id: https://ws10smt.googlecode.com/svn/trunk@393 ec762483-ff6d-05da-a07a-a48fb63a330f
Diffstat (limited to 'decoder/ff.cc')
| -rw-r--r-- | decoder/ff.cc | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/decoder/ff.cc b/decoder/ff.cc index 28d6f732..4f1a3d32 100644 --- a/decoder/ff.cc +++ b/decoder/ff.cc @@ -1,6 +1,6 @@ -//TODO: actually score rule_feature()==true features once only, hash keyed on rule or modify TRule directly? need to keep clear in forest which features come from models vs. rules; then rescoring could drop all the old models features at once +//TODO: non-sparse vector for all feature functions? modelset applymodels keeps track of who has what features? it's nice having FF that could generate a handful out of 10000 possible feats, though. -//TODO: 0 size state != rule-local feature, i.e. still may depend on source span loc/context. identify truly rule-local features so if we want they can be added to grammar rules (minor speedup) +//TODO: actually score rule_feature()==true features once only, hash keyed on rule or modify TRule directly? need to keep clear in forest which features come from models vs. rules; then rescoring could drop all the old models features at once #include <boost/lexical_cast.hpp> #include "ff.h" |