Commit Graph

412 Commits

Author SHA1 Message Date
svlandeg
1de61f68d6 improve speed of prediction loop 2019-06-26 13:53:10 +02:00
svlandeg
bee23cd8af try Tok2Vec instead of SpacyVectors 2019-06-25 16:09:22 +02:00
svlandeg
b58bace84b small fixes 2019-06-24 10:55:04 +02:00
svlandeg
a31648d28b further code cleanup 2019-06-19 09:15:43 +02:00
svlandeg
478305cd3f small tweaks and documentation 2019-06-18 18:38:09 +02:00
svlandeg
0d177c1146 clean up code, remove old code, move to bin 2019-06-18 13:20:40 +02:00
svlandeg
ffae7d3555 sentence encoder only (removing article/mention encoder) 2019-06-18 00:05:47 +02:00
svlandeg
6332af40de baseline performances: oracle KB, random and prior prob 2019-06-17 14:39:40 +02:00
svlandeg
24db1392b9 reprocessing all of wikipedia for training data 2019-06-16 21:14:45 +02:00
svlandeg
81731907ba performance per entity type 2019-06-14 19:55:46 +02:00
svlandeg
b312f2d0e7 redo training data to be independent of KB and entity-level instead of doc-level 2019-06-14 15:55:26 +02:00
svlandeg
0b04d142de regenerating KB 2019-06-13 22:32:56 +02:00
svlandeg
78dd3e11da write entity linking pipe to file and keep vocab consistent between kb and nlp 2019-06-13 16:25:39 +02:00
svlandeg
b12001f368 small fixes 2019-06-12 22:05:53 +02:00
svlandeg
6521cfa132 speeding up training 2019-06-12 13:37:05 +02:00
svlandeg
66813a1fdc speed up predictions 2019-06-11 14:18:20 +02:00
svlandeg
fe1ed432ef eval on dev set, varying combo's of prior and context scores 2019-06-11 11:40:58 +02:00
svlandeg
83dc7b46fd first tests with EL pipe 2019-06-10 21:25:26 +02:00
svlandeg
7de1ee69b8 training loop in proper pipe format 2019-06-07 15:55:10 +02:00
svlandeg
0486ccabfd introduce goldparse.links 2019-06-07 13:54:45 +02:00
svlandeg
a5c061f506 storing NEL training data in GoldParse objects 2019-06-07 12:58:42 +02:00
svlandeg
61f0e2af65 code cleanup 2019-06-06 20:22:14 +02:00
svlandeg
d8b435ceff pretraining description vectors and storing them in the KB 2019-06-06 19:51:27 +02:00
svlandeg
5c723c32c3 entity vectors in the KB + serialization of them 2019-06-05 18:29:18 +02:00
svlandeg
9abbd0899f separate entity encoder to get 64D descriptions 2019-06-05 00:09:46 +02:00
svlandeg
fb37cdb2d3 implementing el pipe in pipes.pyx (not tested yet) 2019-06-03 21:32:54 +02:00
svlandeg
d83a1e3052 Merge branch 'master' into feature/nel-wiki 2019-06-03 09:35:10 +02:00
svlandeg
9e88763dab 60% acc run 2019-06-03 08:04:49 +02:00
svlandeg
268a52ead7 experimenting with cosine sim for negative examples (not OK yet) 2019-05-29 16:07:53 +02:00
svlandeg
a761929fa5 context encoder combining sentence and article 2019-05-28 18:14:49 +02:00
svlandeg
992fa92b66 refactor again to clusters of entities and cosine similarity 2019-05-28 00:05:22 +02:00
svlandeg
8c4aa076bc small fixes 2019-05-27 14:29:38 +02:00
svlandeg
cfc27d7ff9 using Tok2Vec instead 2019-05-26 23:39:46 +02:00
svlandeg
abf9af81c9 learn rate en epochs 2019-05-24 22:04:25 +02:00
svlandeg
86ed771e0b adding local sentence encoder 2019-05-23 16:59:11 +02:00
svlandeg
4392c01b7b obtain sentence for each mention 2019-05-23 15:37:05 +02:00
svlandeg
97241a3ed7 upsampling and batch processing 2019-05-22 23:40:10 +02:00
svlandeg
1a16490d20 update per entity 2019-05-22 12:46:40 +02:00
svlandeg
eb08bdb11f hidden with for encoders 2019-05-21 23:42:46 +02:00
svlandeg
7b13e3d56f undersampling negatives 2019-05-21 18:35:10 +02:00
svlandeg
2fa3fac851 fix concat bp and more efficient batch calls 2019-05-21 13:43:59 +02:00
svlandeg
0a15ee4541 fix in bp call 2019-05-20 23:54:55 +02:00
svlandeg
89e322a637 small fixes 2019-05-20 17:20:39 +02:00
svlandeg
7edb2e1711 fix convolution layer 2019-05-20 11:58:48 +02:00
svlandeg
dd691d0053 debugging 2019-05-17 17:44:11 +02:00
svlandeg
400b19353d simplify architecture and larger-scale test runs 2019-05-17 01:51:18 +02:00
svlandeg
d51bffe63b clean up code 2019-05-16 18:36:15 +02:00
svlandeg
b5470f3d75 various tests, architectures and experiments 2019-05-16 18:25:34 +02:00
svlandeg
9ffe5437ae calculate gradient for entity encoding 2019-05-15 02:23:08 +02:00
svlandeg
2713abc651 implement loss function using dot product and prob estimate per candidate cluster 2019-05-14 22:55:56 +02:00
svlandeg
09ed446b20 different architecture / settings 2019-05-14 08:37:52 +02:00
svlandeg
4142e8dd1b train and predict per article (saving time for doc encoding) 2019-05-13 17:02:34 +02:00
svlandeg
3b81b00954 evaluating on dev set during training 2019-05-13 14:26:04 +02:00
svlandeg
b6d788064a some first experiments with different architectures and metrics 2019-05-10 12:53:14 +02:00
svlandeg
9d089c0410 grouping clusters of instances per doc+mention 2019-05-09 18:11:49 +02:00
svlandeg
c6ca8649d7 first stab at model - not functional yet 2019-05-09 17:23:19 +02:00
svlandeg
9f33732b96 using entity descriptions and article texts as input embedding vectors for training 2019-05-07 16:03:42 +02:00
svlandeg
7e348d7f7f baseline evaluation using highest-freq candidate 2019-05-06 15:13:50 +02:00
Ines Montani
dd153b2b33 Simplify helper (see #3681) [ci skip] 2019-05-06 15:13:10 +02:00
Ines Montani
f8fce6c03c Fix typo (see #3681) 2019-05-06 15:02:11 +02:00
Ines Montani
f2a56c1b56 Rewrite example to use Retokenizer (resolves #3681)
Also add helper to filter spans
2019-05-06 14:51:18 +02:00
svlandeg
6961215578 refactor code to separate functionality into different files 2019-05-06 10:56:56 +02:00
svlandeg
f5190267e7 run only 100M of WP data as training dataset (9%) 2019-05-03 18:09:09 +02:00
svlandeg
4e929600e5 fix WP id parsing, speed up processing and remove ambiguous strings in one doc (for now) 2019-05-03 17:37:47 +02:00
svlandeg
34600c92bd try catch per article to ensure the pipeline goes on 2019-05-03 15:10:09 +02:00
svlandeg
bbcb9da466 creating training data with clean WP texts and QID entities true/false 2019-05-03 10:44:29 +02:00
svlandeg
cba9680d13 run NER on clean WP text and link to gold-standard entity IDs 2019-05-02 17:24:52 +02:00
svlandeg
581dc9742d parsing clean text from WP articles to use as input data for NER and NEL 2019-05-02 17:09:56 +02:00
svlandeg
8353552191 cleanup 2019-05-01 23:26:16 +02:00
svlandeg
1ae41daaa9 allow small rounding errors 2019-05-01 23:05:40 +02:00
svlandeg
3629a52ede reading all persons in wikidata 2019-05-01 01:00:59 +02:00
svlandeg
60b54ae8ce bulk entity writing and experiment with regex wikidata reader to speed up processing 2019-05-01 00:00:38 +02:00
svlandeg
653b7d9c87 calculate entity raw counts offline to speed up KB construction 2019-04-30 11:39:42 +02:00
svlandeg
19e8f339cb deduce entity freq from WP corpus and serialize vocab in WP test 2019-04-29 17:37:29 +02:00
svlandeg
54d0cea062 unit test for KB serialization 2019-04-24 23:52:34 +02:00
svlandeg
3e0cb69065 KB aliases to and from file 2019-04-24 20:24:24 +02:00
svlandeg
ad6c5e581c writing and reading number of entries to/from header 2019-04-24 15:31:44 +02:00
svlandeg
6e3223f234 bulk loading in proper order of entity indices 2019-04-24 11:26:38 +02:00
svlandeg
694fea597a dumping all entryC entries + (inefficient) reading back in 2019-04-23 18:36:50 +02:00
svlandeg
8e70a564f1 custom reader and writer for _EntryC fields (first stab at it - not complete) 2019-04-23 16:33:40 +02:00
svlandeg
004e5e7d1c little fixes 2019-04-19 14:24:02 +02:00
svlandeg
9a8197185b fix alias capitalization 2019-04-18 22:37:50 +02:00
svlandeg
9f308eb5dc fixes for prior prob and linking wikidata IDs with wikipedia titles 2019-04-18 16:14:25 +02:00
svlandeg
10ee8dfea2 poc with few entities and collecting aliases from the WP links 2019-04-18 14:12:17 +02:00
svlandeg
6763e025e1 parse wp dump for links to determine prior probabilities 2019-04-15 11:41:57 +02:00
svlandeg
3163331b1e wikipedia dump parser and mediawiki format regex cleanup 2019-04-14 21:52:01 +02:00
svlandeg
b31a390a9a reading types, claims and sitelinks 2019-04-11 21:42:44 +02:00
svlandeg
6e997be4b4 reading wikidata descriptions and aliases 2019-04-11 21:08:22 +02:00
svlandeg
9a7d534b1b enable nogil for cython functions in kb.pxd 2019-04-10 17:25:10 +02:00
Ines Montani
24cecdb44f Update compatibility [ci skip] 2019-04-01 16:25:16 +02:00
Sofie
a4a6bfa4e1
Merge branch 'master' into feature/el-framework 2019-03-26 11:00:02 +01:00
svlandeg
8814b9010d entity as one field instead of both ID and name 2019-03-25 18:10:41 +01:00
Matthew Honnibal
6c783f8045 Bug fixes and options for TextCategorizer (#3472)
* Fix code for bag-of-words feature extraction

The _ml.py module had a redundant copy of a function to extract unigram
bag-of-words features, except one had a bug that set values to 0.
Another function allowed extraction of bigram features. Replace all three
with a new function that supports arbitrary ngram sizes and also allows
control of which attribute is used (e.g. ORTH, LOWER, etc).

* Support 'bow' architecture for TextCategorizer

This allows efficient ngram bag-of-words models, which are better when
the classifier needs to run quickly, especially when the texts are long.
Pass architecture="bow" to use it. The extra arguments ngram_size and
attr are also available, e.g. ngram_size=2 means unigram and bigram
features will be extracted.

* Fix size limits in train_textcat example

* Explain architectures better in docs
2019-03-23 16:44:44 +01:00
svlandeg
9de9900510 adding future import unicode literals to .py files 2019-03-22 16:18:04 +01:00
Matthew Honnibal
4c5f265884
Fix train loop for train_textcat example 2019-03-22 16:10:11 +01:00
svlandeg
5318ce88fa 'entity_linker' instead of 'el' 2019-03-22 13:55:10 +01:00
svlandeg
a48241e9a2 use nlp's vocab for stringstore 2019-03-22 11:36:45 +01:00
svlandeg
1ee0e78fd7 select candidate with highest prior probabiity 2019-03-22 11:36:45 +01:00
Matthew Honnibal
4e3ed2ea88 Add -t2v argument to train_textcat script 2019-03-20 23:05:42 +01:00
Ines Montani
399987c216 Test and update examples [ci skip] 2019-03-16 14:15:49 +01:00