summaryrefslogtreecommitdiff
path: root/decoder/hg.h
AgeCommit message (Collapse)Author
2010-08-16greedy binarization - needs testing, may have broke l2rgraehl@gmail.com
git-svn-id: https://ws10smt.googlecode.com/svn/trunk@560 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-08-13(NEEDS TESTING) cfg index rules->nts, sort by prob, remove duplicates ↵graehl
keeping highest prob, topo sort (and after binarize topo sort). beginning to apply_fsa_models (PrefixTrieNode) git-svn-id: https://ws10smt.googlecode.com/svn/trunk@539 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-08-11major refactor, break bad circular depsredpony
git-svn-id: https://ws10smt.googlecode.com/svn/trunk@509 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-08-10NTSpan set l,rgraehl
git-svn-id: https://ws10smt.googlecode.com/svn/trunk@502 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-08-10cdec --cfg_output=-graehl@gmail.com
git-svn-id: https://ws10smt.googlecode.com/svn/trunk@499 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-08-09hg sort in_edges by viterbi, debug print temporarily enabled which always ↵graehl@gmail.com
--show_derivation git-svn-id: https://ws10smt.googlecode.com/svn/trunk@494 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-08-08cfggraehl@gmail.com
git-svn-id: https://ws10smt.googlecode.com/svn/trunk@493 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-08-02fake tdict names for non-ids, push viterbi cost to root in hg, store as ↵graehl@gmail.com
feature. type erased fsa feature via virtual interface. made lexical_cast assume C locale for speed. git-svn-id: https://ws10smt.googlecode.com/svn/trunk@465 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-07-28debugging print - still no idea on .05% difference scoring 3gram using phrasesgraehl
git-svn-id: https://ws10smt.googlecode.com/svn/trunk@446 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-07-26fix name collisionredpony
git-svn-id: https://ws10smt.googlecode.com/svn/trunk@423 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-07-26comment cleanupgraehl
git-svn-id: https://ws10smt.googlecode.com/svn/trunk@421 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-07-26stateful ff_from_fsa works - fixed bug that never moved right state from ↵graehl
rightmost variable up to result git-svn-id: https://ws10smt.googlecode.com/svn/trunk@413 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-07-25USE_INFO_EDGE 0 includegraehl
git-svn-id: https://ws10smt.googlecode.com/svn/trunk@412 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-07-25USE_INFO_EDGE works for exh, cubegraehl
git-svn-id: https://ws10smt.googlecode.com/svn/trunk@411 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-07-25being ensuring edge debug info is copied when apply_models - new hg.AddEdge ↵graehl
methods git-svn-id: https://ws10smt.googlecode.com/svn/trunk@409 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-07-25fixed CreateViterbiHypergraph (old impl did not work), so --show_derivation ↵graehl
works git-svn-id: https://ws10smt.googlecode.com/svn/trunk@408 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-07-25more comprehensible (but untested) edge/node filtering, awesome per-edge ↵graehl
debugging streams git-svn-id: https://ws10smt.googlecode.com/svn/trunk@407 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-07-25cleaned up kbest, new USE_INFO_EDGE 1 logs per edge, --show_derivation ↵graehl
(needs work; handle kbest deriv, viterbi deriv, sort hg exposing viterbi?) git-svn-id: https://ws10smt.googlecode.com/svn/trunk@405 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-07-24FSA: simpler Scan1 ScanT1 methods, otherewise also expose edge to full Scangraehl
git-svn-id: https://ws10smt.googlecode.com/svn/trunk@399 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-07-16smallvector templategraehl
git-svn-id: https://ws10smt.googlecode.com/svn/trunk@302 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-07-15Fix bug in Density/BeamPruneInsideOutsideadam.d.lopez
-only affected code calling these (legacy) functions, which have been refactored. -DensityPIO implemented BeamPIO and vice versa. git-svn-id: https://ws10smt.googlecode.com/svn/trunk@266 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-07-12DEBUG_PROMISE - looks okgraehl
git-svn-id: https://ws10smt.googlecode.com/svn/trunk@220 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-07-12inebriated, but sped up inside/outside pruning and made cube poplimit scale ↵graehl
with promise of nodes from previous global pruning, if --promise_power=N for N>0. git-svn-id: https://ws10smt.googlecode.com/svn/trunk@219 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-07-08--show_featuresgraehl
git-svn-id: https://ws10smt.googlecode.com/svn/trunk@184 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-07-07safe hg pruning without needing additional inside reachability pass (max ↵graehl
margin tightness is less at bottom of derivation tree) git-svn-id: https://ws10smt.googlecode.com/svn/trunk@181 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-07-06prune epsilon, comments, use inside pruning for probably no segfaults in vestgraehl
git-svn-id: https://ws10smt.googlecode.com/svn/trunk@167 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-07-01factor forest stats (show size post pruning, and portion kept) and cdec ↵graehl
--scale_prune_srclen git-svn-id: https://ws10smt.googlecode.com/svn/trunk@96 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-06-24build on cygwingraehl
git-svn-id: https://ws10smt.googlecode.com/svn/trunk@23 ec762483-ff6d-05da-a07a-a48fb63a330f
2010-06-22initial checkinredpony
git-svn-id: https://ws10smt.googlecode.com/svn/trunk@2 ec762483-ff6d-05da-a07a-a48fb63a330f