diff options
| author | graehl <graehl@ec762483-ff6d-05da-a07a-a48fb63a330f> | 2010-07-06 22:06:26 +0000 |
|---|---|---|
| committer | graehl <graehl@ec762483-ff6d-05da-a07a-a48fb63a330f> | 2010-07-06 22:06:26 +0000 |
| commit | 37070233d8768919bc70fa9bd8a33e04fbe661d3 (patch) | |
| tree | ee5df3f276de4ce47432b0f7e9ee085600285eb3 /decoder/hg.h | |
| parent | 9f2f8c40c1e074ba08119dfb9ed55f278d6a51f2 (diff) | |
prune epsilon, comments, use inside pruning for probably no segfaults in vest
git-svn-id: https://ws10smt.googlecode.com/svn/trunk@167 ec762483-ff6d-05da-a07a-a48fb63a330f
Diffstat (limited to 'decoder/hg.h')
| -rw-r--r-- | decoder/hg.h | 1 |
1 files changed, 1 insertions, 0 deletions
diff --git a/decoder/hg.h b/decoder/hg.h index a632dc1c..e54b4bcc 100644 --- a/decoder/hg.h +++ b/decoder/hg.h @@ -166,6 +166,7 @@ class Hypergraph { // remove edges from the hypergraph if prune_edge[edge_id] is true // TODO need to investigate why this shouldn't be run for the forest trans // case. To investigate, change false to true and see where ftrans crashes + //TODO: what does the above "TODO" comment mean? that PruneEdges can lead to a crash? or that run_inside_algorithm should be false? there definitely is an unsolved bug, see hg.cc - workaround added void PruneEdges(const std::vector<bool>& prune_edge, bool run_inside_algorithm = false); // for density>=1.0, keep this many times the edges needed for the 1best derivation |