explosion/spaCy
1
1
Fork 0
mirror of https://github.com/explosion/spaCy.git synced 2024-11-13 13:17:06 +03:00
Code Issues Packages Projects Releases Wiki Activity
9,064 Commits 68 Branches 175 Tags 279 MiB
93be3ad038
Commit Graph

349 Commits

Author SHA1 Message Date
Matthew Honnibal
008e1ee1dd Update pretrain command 2018-11-29 12:36:43 +00:00
Matthew Honnibal
61e435610e
💫 Feature/improve pretraining (#2971) 
* Improve spacy pretrain script

* Implement BERT-style 'masked language model' objective. Much better
results.

* Improve logging.

* Add length cap for documents, to avoid memory errors.

* Require thinc 7.0.0.dev1

* Require thinc 7.0.0.dev1

* Add argument for using pretrained vectors

* Fix defaults

* Fix syntax error

* Improve spacy pretrain script

* Implement BERT-style 'masked language model' objective. Much better
results.

* Improve logging.

* Add length cap for documents, to avoid memory errors.

* Require thinc 7.0.0.dev1

* Require thinc 7.0.0.dev1

* Add argument for using pretrained vectors

* Fix defaults

* Fix syntax error

* Tweak pretraining script

* Fix data limits in spacy.gold

* Fix pretrain script
 2018-11-28 18:04:58 +01:00
Matthew Honnibal
ef0820827a
Update hyper-parameters after NER random search (#2972) 
These experiments were completed a few weeks ago, but I didn't make the PR, pending model release.

    Token vector width: 128->96
    Hidden width: 128->64
    Embed size: 5000->2000
    Dropout: 0.2->0.1
    Updated optimizer defaults (unclear how important?)

This should improve speed, model size and load time, while keeping
similar or slightly better accuracy.

The tl;dr is we prefer to prevent over-fitting by reducing model size,
rather than using more dropout.
 2018-11-27 18:49:52 +01:00
Ines Montani
b4581435f6 Merge branch 'develop' of https://github.com/explosion/spaCy into develop 2018-11-16 13:08:22 +01:00
Ines Montani
e2f75eb492 Fix message formatting 2018-11-16 13:08:20 +01:00
Matthew Honnibal
2874b8efd8 Fix tok2vec loading in spacy train 2018-11-15 23:34:54 +00:00
Matthew Honnibal
2ddd428834 Fix pretrain script 2018-11-15 23:34:35 +00:00
Matthew Honnibal
f8afaa0c1c Fix pretrain 2018-11-15 22:46:53 +00:00
Matthew Honnibal
6af6950e46 Fix pretrain 2018-11-15 22:45:36 +00:00
Matthew Honnibal
3e7b214e57 Make pretrain script work with stream from stdin 2018-11-15 22:44:07 +00:00
Matthew Honnibal
8fdb9bc278
💫 Add experimental ULMFit/BERT/Elmo-like pretraining (#2931) 
* Add 'spacy pretrain' command

* Fix pretrain command for Python 2

* Fix pretrain command

* Fix pretrain command
 2018-11-15 22:17:16 +01:00
Matthew Honnibal
8f2a6367e9 Fix usage of PyTorch BiLSTM in ud_train 2018-09-13 22:54:59 +00:00
Matthew Honnibal
445b81ce3f Support bilstm_depth argument in ud-train 2018-09-13 19:30:22 +02:00
Matthew Honnibal
3eb9f3e2b8 Fix defaults for ud-train 2018-09-13 18:05:48 +02:00
Matthew Honnibal
59cf533879 Improve ud-train script. Make config optional 2018-09-13 14:24:08 +02:00
Matthew Honnibal
da7650e84b Fix maximum doc length in ud_train script 2018-09-13 14:10:25 +02:00
Matthew Honnibal
4d2d7d5866 Fix new feature flags 2018-08-27 02:12:39 +02:00
Matthew Honnibal
9c33d4d1df Add more hyper-parameters to spacy ud-train 
* subword_features: Controls whether subword features are used in the
word embeddings. True by default (specifically, prefix, suffix and word
shape). Should be set to False for languages like Chinese and Japanese.

* conv_depth: Depth of the convolutional layers. Defaults to 4.
 2018-08-27 01:48:46 +02:00
Matthew Honnibal
595c893791 Expose noise_level option in train CLI 2018-08-16 00:41:44 +02:00
Matthew Honnibal
6ea981c839 Add converter for jsonl NER data 2018-08-14 14:04:32 +02:00
Matthew Honnibal
02c5c114d0 Fix usage of deprecated freqs.txt in init-model 2018-08-14 13:19:15 +02:00
Matthew Honnibal
4336397ecb Update develop from master 2018-08-14 03:04:28 +02:00
Xiaoquan Kong
f0c9652ed1 New Feature: display more detail when Error E067 (#2639) 
* Fix off-by-one error

* Add verbose option

* Update verbose option

* Update documents for verbose option
 2018-08-07 10:45:29 +02:00
Kaisa (Katarzyna) Korsak
e531a827db Changed conllu2json to be able to extract NER tags (#2594) 
* extract ner tags from conllu file if available

* fixed a bug in regex
 2018-07-25 22:21:31 +02:00
ines
d84b13e02c Merge branch 'master' into develop 2018-07-18 18:57:00 +02:00
Ole Henrik Skogstrøm
6e2930a4a2 Conll(u)-bio converter (#2525) 
* Started simple conllxbiluo converter

* Fix missing BIO to BILUO conversion
 2018-07-18 18:55:42 +02:00
Matthew Honnibal
8ae1bec8bf Fix init_model 2018-07-05 14:02:06 +02:00
Matthew Honnibal
dee8bdb900 Fix init-model for npz vectors 2018-07-04 02:29:48 +02:00
Matthew Honnibal
59d655e8d0 Fix model init from jsonl 2018-07-04 01:30:40 +02:00
Matthew Honnibal
1e38bea6e9 Save vectors init 2018-07-03 23:55:04 +02:00
Matthew Honnibal
6692833887 Fix init_model 2018-07-03 23:24:11 +02:00
Matthew Honnibal
4a38a26cb5 Fix init_model 2018-07-03 22:57:11 +02:00
Matthew Honnibal
019d09e3c3 Fix init model 2018-07-03 22:16:44 +02:00
Matthew Honnibal
2543f8c93a Support .npz vectors in init-model command 2018-07-03 21:42:16 +02:00
Matthew Honnibal
86aad11939 Fix init_model arg 2018-07-03 17:00:42 +02:00
Matthew Honnibal
eff42d36e3 Fix init model command 2018-07-03 16:32:23 +02:00
Matthew Honnibal
6a89faf12e Add support for jsonl-formatted lexical attributes to init-model command. 2018-07-03 12:22:56 +02:00
Matthew Honnibal
c83fccfe2a Fix output of best model 2018-06-25 23:05:56 +02:00
Matthew Honnibal
69c900f003 Fix init-model if no vectors provided 2018-06-25 18:26:02 +02:00
Matthew Honnibal
664f89327a Fix init-model if no vectors provided 2018-06-25 17:58:45 +02:00
Matthew Honnibal
c4698f5712 Don't collate model unless training succeeds 2018-06-25 16:36:42 +02:00
Matthew Honnibal
24dfbb8a28 Fix model collation 2018-06-25 14:35:24 +02:00
Matthew Honnibal
62237755a4 Import shutil 2018-06-25 13:40:17 +02:00
Matthew Honnibal
a040fca99e Import json into cli.train 2018-06-25 11:50:37 +02:00
Matthew Honnibal
2c703d99c2 Fix collation of best models 2018-06-25 01:21:34 +02:00
Matthew Honnibal
2c80b7c013 Collate best model after training 2018-06-24 23:39:52 +02:00
ines
330c039106 Merge branch 'master' into develop 2018-05-26 18:30:52 +02:00
James Messinger
4515e96e90 Better formatting for spacy train CLI (#2357) 
* Better formatting for `spacy train` CLI

Changed to use fixed-spaces rather than tabs to align table headers and data.

### Before:
```
Itn.    P.Loss  N.Loss  UAS     NER P.  NER R.  NER F.  Tag %   Token %
0       4618.857        2910.004        76.172  79.645  67.987  88.732  88.261  100.000 4436.9  6376.4
1       4671.972        3764.812        74.481  78.046  62.374  82.680  88.377  100.000 4672.2  6227.1
2       4742.756        3673.473        71.994  77.380  63.966  84.494  90.620  100.000 4298.0  5983.9
```

### After:
```
Itn.  Dep Loss  NER Loss  UAS     NER P.  NER R.  NER F.  Tag %   Token %  CPU WPS  GPU WPS
0     4618.857  2910.004  76.172  79.645  67.987  88.732  88.261  100.000  4436.9   6376.4
1     4671.972  3764.812  74.481  78.046  62.374  82.680  88.377  100.000  4672.2   6227.1
2     4742.756  3673.473  71.994  77.380  63.966  84.494  90.620  100.000  4298.0   5983.9
```

* Added contributor file
 2018-05-25 13:08:45 +02:00
Matthew Honnibal
ce458c2428 Fix spacy requirement constraint in package template 2018-05-22 20:50:46 +02:00
Matthew Honnibal
f3b4f6a4ec Merge setup.py 2018-05-20 23:21:00 +02:00