Commit Graph

  • 9b13d11ab3 * Fix handling of entities in StateClass Matthew Honnibal 2015-06-16 23:35:21 +0200
  • 5699585278 * Use tree_arc_eager system as baseline in experiments Matthew Honnibal 2015-06-15 08:23:43 +0200
  • c40a2c661c * Add tree_arc_eager Matthew Honnibal 2015-06-15 08:23:24 +0200
  • a5ae98a543 * Add tree_arc_eager to setup.py Matthew Honnibal 2015-06-15 08:22:59 +0200
  • 5da5cf7084 * Add some more features for S1/S0 Matthew Honnibal 2015-06-15 04:07:13 +0200
  • 8156a01bca * Fix root label for orig_arc_eager Matthew Honnibal 2015-06-15 02:54:55 +0200
  • 21930ede15 * Switch toggle on USE_ROOT_ARC_SEGMENT Matthew Honnibal 2015-06-15 02:54:32 +0200
  • 4841f8ad5e * Set transition system early Matthew Honnibal 2015-06-15 02:54:12 +0200
  • 38a6afa484 * Make possibly dubious correction to the unshift oracle Matthew Honnibal 2015-06-15 02:50:00 +0200
  • f66228f253 * Add some more features, esp for labels Matthew Honnibal 2015-06-14 21:17:39 +0200
  • 3da8e0f317 * Add orig_arc_eager Matthew Honnibal 2015-06-14 20:31:44 +0200
  • bcfdf126a4 * Add toggle for OrigArcEager system Matthew Honnibal 2015-06-14 20:28:14 +0200
  • ea8a103007 * Fix import of TransitionSystem in parser.pyx Matthew Honnibal 2015-06-14 19:01:26 +0200
  • e0984ca139 * Fix valency features in StateClass Matthew Honnibal 2015-06-14 17:50:26 +0200
  • e50ac1a47f * Add verbose printing to scorer Matthew Honnibal 2015-06-14 17:45:50 +0200
  • c500d72dc2 * Temporarily disable NER, and wire up the verbose flag during training Matthew Honnibal 2015-06-14 17:45:31 +0200
  • 763cbd23d5 * Upd stateclass.print_state Matthew Honnibal 2015-06-14 17:44:29 +0200
  • bdd07bf000 * Fix Break oracle, but disable the Break transition for now, while we finalize the gold-standard experiments Matthew Honnibal 2015-06-14 17:44:03 +0200
  • 399f15fbdf * Add flag to toggle handling of multi-root inputs without the Break transition. Clear up now unused best_valid stuff. Matthew Honnibal 2015-06-14 00:28:37 +0200
  • 75289b4761 * Don't refuse to parse single token sentences, incase some transition system needs them, e.g. single word entity. Instead fix error in _init_state. Matthew Honnibal 2015-06-13 22:55:55 +0200
  • 77d7e79c7e * Fix r/l and distance features. Matthew Honnibal 2015-06-12 13:06:15 +0200
  • b643cb3d5c * Allow training documents to be filtered in gold.pyx Matthew Honnibal 2015-06-12 02:42:08 +0200
  • 15e177d7a1 * Fixes to unshift/fast-forward strategy. Getting 91.55 greedy on NW dev, gold preproc Matthew Honnibal 2015-06-12 01:50:23 +0200
  • afd77a529b * Prepare for break transition, with fast-forwarding. 86.5 on 1k nw gold preproc Matthew Honnibal 2015-06-10 14:08:30 +0200
  • 495f528709 * Add support for sentence breaks in stateclass Matthew Honnibal 2015-06-10 12:33:55 +0200
  • b7b18c279d * Fix Reduce oracle. Getting 86.35 Matthew Honnibal 2015-06-10 11:33:39 +0200
  • bb09b5d91a * Fix shifted bit vector in stateclass --- should reflect whether the word has been *unshifted*. Matthew Honnibal 2015-06-10 11:33:09 +0200
  • aa9625f688 * Do non-monotonic Unshift. Every word can be shifted at most 1 time. When the Reduce move is used, if S0 has no head, we put the word back on the buffer. Gets 86.4 on nw 1k with gold pre-proc. Break transition not yet implemented for this. Matthew Honnibal 2015-06-10 10:15:56 +0200
  • 7bf6b7de3e * Add unshift action to StateClass, and track which moves have been shifted Matthew Honnibal 2015-06-10 10:13:03 +0200
  • f7c8069e65 * Fix bug in distance feature Matthew Honnibal 2015-06-10 10:12:17 +0200
  • abd07c067a * Inline B and S methods on stateclass Matthew Honnibal 2015-06-10 07:22:33 +0200
  • e2f9a80713 * Remove old _state imports Matthew Honnibal 2015-06-10 07:09:17 +0200
  • e9aaecc619 * Remove from_struct method from StateClass Matthew Honnibal 2015-06-10 06:58:27 +0200
  • 18cc326dc0 * Bug fixes to ner.pyx Matthew Honnibal 2015-06-10 06:57:41 +0200
  • 90a3add8d7 * Require thinc 2.0 Matthew Honnibal 2015-06-10 06:57:13 +0200
  • e5570c9700 * Set nogil for oracle functions Matthew Honnibal 2015-06-10 06:56:35 +0200
  • 4575e7a60f * Fix beam search with new StateClass Matthew Honnibal 2015-06-10 06:33:39 +0200
  • d70304b7dd * Require newer thinc Matthew Honnibal 2015-06-10 04:20:42 +0200
  • 04b1cd9b8c * Greedy parsing working with new StateClass. Beam parsing broken Matthew Honnibal 2015-06-10 04:20:23 +0200
  • 6a94b64eca * Remove State* from parser.pyx entirely, switching over to StateClass. Beam parsing still untested. Matthew Honnibal 2015-06-10 02:03:38 +0200
  • f14a1526aa * Remove version of fill_context that takes State* Matthew Honnibal 2015-06-10 01:39:07 +0200
  • d68c686ec1 * Move StateClass into interface of transition functions Matthew Honnibal 2015-06-10 01:35:28 +0200
  • 4b98b3e9c8 * Cost functions now take StateClass argument, instead of State*. Matthew Honnibal 2015-06-10 00:40:43 +0200
  • e0cf61f591 * Move StateClass into the interface for is_valid Matthew Honnibal 2015-06-09 23:23:28 +0200
  • 09617a4638 * Whitespace Matthew Honnibal 2015-06-09 21:20:33 +0200
  • 0895d454fb * Prepare to switch to using state class, instead of state struct Matthew Honnibal 2015-06-09 21:20:14 +0200
  • 2b9629ed62 * Begin adding stateclass to ArcEager Matthew Honnibal 2015-06-09 01:41:09 +0200
  • ba10fd8af5 * Add StateClass, to replace/refactor the mess in _state Matthew Honnibal 2015-06-09 01:39:54 +0200
  • bd4f5f89cb * Add note about failed tokenization Matthew Honnibal 2015-06-08 16:17:07 +0200
  • c7e3dfc1dc * Don't automatically push words when stack is empty, as it messes up beam parsing. Add hash method to beam state. Matthew Honnibal 2015-06-08 14:49:04 +0200
  • d51a86478e * Upd contact information on license page Matthew Honnibal 2015-06-08 02:20:31 +0200
  • be9ccdac6d * Upd docs Matthew Honnibal 2015-06-08 02:17:52 +0200
  • e46cef634f * Update docs Matthew Honnibal 2015-06-08 02:01:44 +0200
  • 2aadb687c3 * Note new release in docs Matthew Honnibal 2015-06-08 01:47:06 +0200
  • d318cf1497 * Have travis use pip again... Matthew Honnibal 2015-06-08 01:27:08 +0200
  • 7cd0697e64 * Avoid downloading data in travis.yml --- just test the non-parser, non-vectors modules. Matthew Honnibal 2015-06-08 00:57:19 +0200
  • 00a0dfcb59 * Avoid shipping the spacy.munge package Matthew Honnibal 2015-06-08 00:54:13 +0200
  • 7d265a9c62 * Revert to wget in spacy.en.download Matthew Honnibal 2015-06-08 00:48:56 +0200
  • a8fc5f1285 * Fix munge/read_ner Matthew Honnibal 2015-06-08 00:35:04 +0200
  • 1515862861 * Fix download.py Matthew Honnibal 2015-06-08 00:08:05 +0200
  • 7e9e8f654a * Use urllib in spacy.en.download Matthew Honnibal 2015-06-07 23:51:38 +0200
  • 514d6b8e22 * Upd travis.yml Matthew Honnibal 2015-06-07 23:26:41 +0200
  • 1d26f77412 * Upd travis.yml Matthew Honnibal 2015-06-07 23:19:42 +0200
  • dcdf566ce3 * Fix fab test Matthew Honnibal 2015-06-07 22:59:05 +0200
  • 68be03d66b * Fiddle with travis.yml Matthew Honnibal 2015-06-07 22:57:48 +0200
  • 22f1ad012e * Add spacy.munge to list of packages Matthew Honnibal 2015-06-07 22:28:13 +0200
  • ce8e524825 * Fix requirements in setup.py Matthew Honnibal 2015-06-07 22:24:21 +0200
  • 7195c07897 * Add explicit tokenizer test Matthew Honnibal 2015-06-07 21:29:17 +0200
  • 957e6eaa8a * Fixes to prag_sbd tests Matthew Honnibal 2015-06-07 21:19:17 +0200
  • 7a9e9ab9b2 * Update prag_sbd challenge set Matthew Honnibal 2015-06-07 21:13:55 +0200
  • 993128032f * Fiddle with travis.yml Matthew Honnibal 2015-06-07 20:57:18 +0200
  • ddb11f5585 * Fiddle with travis.yml Matthew Honnibal 2015-06-07 20:47:45 +0200
  • e85d1a3808 * Fix test_conjuncts Matthew Honnibal 2015-06-07 19:31:16 +0200
  • 80cff41a9c * Upd download.py Matthew Honnibal 2015-06-07 19:13:28 +0200
  • 6e2564239d * Bug fixes to beam parser. Search still broken on non-gold sentences Matthew Honnibal 2015-06-07 19:12:59 +0200
  • 1ec4e6fc95 * Don't score whitespace tokens Matthew Honnibal 2015-06-07 19:10:32 +0200
  • de8f32ba4d * Upd version in docs Matthew Honnibal 2015-06-07 19:09:39 +0200
  • 731e5f1e46 * Add get() function in spacy/syntax/Config Matthew Honnibal 2015-06-07 19:09:15 +0200
  • ac422492cf * Fix write_parses mode of bin/parser/train.py Matthew Honnibal 2015-06-07 19:08:48 +0200
  • 1cfa326f6e * Comment out test_conjuncts Matthew Honnibal 2015-06-07 19:08:04 +0200
  • 48bc4122d8 * Upd version in setup.py Matthew Honnibal 2015-06-07 19:05:28 +0200
  • 638e07939d * Avoid laoding vectors in test_token_references Matthew Honnibal 2015-06-07 19:03:16 +0200
  • 50768241b3 * Fix test_docs.py Matthew Honnibal 2015-06-07 19:02:43 +0200
  • d83255db17 * Fix ner test Matthew Honnibal 2015-06-07 18:57:42 +0200
  • c6dc151fc3 * Fix spans/test_merge.py Matthew Honnibal 2015-06-07 18:46:16 +0200
  • 2676240cbb * Fix spans/test_merge.py Matthew Honnibal 2015-06-07 18:45:19 +0200
  • 9abb0dd4fd * Fix spans/test_merge.py Matthew Honnibal 2015-06-07 18:44:18 +0200
  • 8a4c9c33f1 * Fix test_token_references test Matthew Honnibal 2015-06-07 18:33:04 +0200
  • 15123329b1 * Have travis test the pip version of the code Matthew Honnibal 2015-06-07 18:17:19 +0200
  • 5f44adc659 * Add tests/spans/conftest.py Matthew Honnibal 2015-06-07 18:07:59 +0200
  • dd587b7477 * Fix tests Matthew Honnibal 2015-06-07 18:07:32 +0200
  • e3af6af83c * Add tests/vocab/conftest.py Matthew Honnibal 2015-06-07 18:02:47 +0200
  • 88041f69d1 * More work on reorganising tests, using conftest.py Matthew Honnibal 2015-06-07 18:02:24 +0200
  • 674ee5dde7 * Add conftest.py to tests/, to allow session-global pipeline. This allows much faster tests. Matthew Honnibal 2015-06-07 17:53:14 +0200
  • 877abb0e5b * Set up tokenizer/ tests properly, using a session-scoped fixture to avoid long load/unload times. Tokenizer tests now complete in 20 seconds. Matthew Honnibal 2015-06-07 17:24:49 +0200
  • 1d5f20fdda * Move nlp variable from global scope Matthew Honnibal 2015-06-07 16:55:11 +0200
  • d37dca72dd * Reorganize tests Matthew Honnibal 2015-06-07 16:49:46 +0200
  • 2ef3555d88 * Add ujson to requirements.txt Matthew Honnibal 2015-06-07 03:22:17 +0200
  • 8f142c1838 * Refactor transition system oracles, to split out move and label cost. Preparing to add Unshift move. Will exclude non-monotonic. Matthew Honnibal 2015-06-07 03:21:29 +0200
  • e2578fbb90 * Avoid parsing and tagging in test_emoticons Matthew Honnibal 2015-06-06 05:59:20 +0200