Skip to content

Reusable #22

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Open
duncanka wants to merge 89 commits into
base: easy-to-use
Choose a base branch
from
Open

Reusable #22

duncanka wants to merge 89 commits into from

Conversation

@duncanka
Copy link
Contributor

@duncanka duncanka commented Sep 17, 2017 •
edited
Loading

  • Improved code structure that is easier to adapt to other neural transition tagging tasks
  • Made it possible for CNN library to release memory (necessary for repeatedly creating networks in the same run)
  • Fixed command line documentation

I've confirmed that the pretrained model still gets a UAS of 93.55% / LAS of 92.16% on the test set.

duncanka added 30 commits January 15, 2017 11:32
@duncanka
Minor style & documentation updates
5f93c72
@duncanka
Made TrainingCorpus an intermediate level in the hierarchy
51cda20
@duncanka
Unified sentence-reading for oracle transition readers
4423e46
@duncanka
Fixed build settings to be better cmake style
5e27f5b
@duncanka
Split up transition-reading code
1514d35 
Shortens function; also allows it to be reused for other
transition-based NLP systems.
@duncanka
Moved string replacement function
a49a559
@duncanka
Moved recording functions into base training corpus reader
39d0efc 
Also some corrections for consistent variable naming conventions
@duncanka
Style fix
40ec8b8
@duncanka
Indentation fix
0401383
@duncanka
Cleaned up oracle action recording a bit
dfd6a6e
@duncanka
Switched action-to-label to be more general and use boost::regex
1991a75
@duncanka
Removed unnecessary record fn params; added vocab copy ctor
566cf57
@duncanka
Fixed training command in README
96fadfc
@duncanka
Correct-count pointer for logprob defaults to null
9fc237d
@duncanka
Minor variable name cleanup for consistency
651a3ca
@duncanka
Allowed accessing final parser state outside LogProbParser
e6afc73
@duncanka
Fixed command line processing to output help if no args given
3f77fcf
@duncanka
More naming consistency
6975c90
@duncanka
Got rid of meaningless 'err' reporting when evaluating
0f05ecd
@duncanka
Added logprob from NN to ParseTree data structure
01736cb 
Makes numbers in logging accurate
@duncanka
Some typedefs
c693089
@duncanka
Switched everything over to Sentence objects
a77168d
@duncanka
Replaced some map find() calls with at() calls
6a90c5e
@duncanka
Removed old comment
014cd26
@duncanka
Added checks for files being provided and successfully opened
95ab82f 
Also added kUNK to the corpus class
@duncanka
Abstracted out much of the LSTM transition tagger architecture
0b9684b 
(In preparation for reusing the same architecture for another
tagger that is not a parser.)
@duncanka
Fixed handling of -w flag
ee8966e
@duncanka
Lint
f8c2aca
@duncanka
A bit of rearranging within the inheritance hierarchy
1b49452
@duncanka
Updated ShouldTerminate interface
818eac8
duncanka added 30 commits March 18, 2017 22:59
@duncanka
A bit of cleanup
e27f330
@duncanka
Minor memory management improvement
5a22567
@duncanka
Possible minor memory management improvement
478e3ff
@duncanka
Made main LogProbTagger public
9c2be25
@duncanka
Minor parameter order change
4e470b9
@duncanka
Sensible default params for LogProbTagger
24d9e5c
@duncanka
Don't use reference actions in test, even if they're specified
88e1af0
@duncanka
*Do* still update the correct count if applicable, even in dev
c97a0f4
@duncanka
Can now shrink a memory pool back down w/o clearing entirely
0e065e0
@duncanka
Attempted to get GPU compilation working
3662e8e
@duncanka
Unused final parser state -> expose arbitrary network states
ae6f7c9
@duncanka
Switched to map for states to expose
43dd29a
@duncanka
Code formatting
7a02eb5
@duncanka
Expose tree node embeddings
9124b54
@duncanka
Exposed whether tagger is in training to subclass functions
93e5708
@duncanka
Minor formatting
bfcea60
@duncanka
Added optional ParseTree pointer to Sentence
5619f81
@duncanka
Fixed assertion bug
d7902d9
@duncanka
Added num_values() to CNN LookupParameters
5653ab7
@duncanka
Made all TaggerStates modifiable by overriding subclass member fns
372cdea
@duncanka
Made Sentences swappable (also more efficient GetWord default)
c8d7685
@duncanka
Fixed Sentence swap function
8b6d315
@duncanka
Updated test command line in README to latest flags
7d07464
@duncanka
Merge branch 'easy-to-use' into reusable
d62ddb8
@duncanka
Deleted dead variable declaration
741dad6
@duncanka
Enabled GetActionProbabilities to signal that oracle action should be…
a4670be 
… used
@duncanka
Allowed attaching metadata to Sentence objects
53756b2
@duncanka
Fixed Sentence printer for non-training corpora
13f4e6b
@duncanka
Improved sentence metadata handling; create dir for model if needed
be9091d
@duncanka
Updated WordForToken to allow iterator hinting for speed
a951527
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@duncanka