Update README.md
Browse files
README.md
CHANGED
@@ -9,11 +9,11 @@ datasets:
|
|
9 |
inference: false
|
10 |
---
|
11 |
|
12 |
-
##
|
13 |
|
14 |
-
This is the standard 4-class NER model for
|
15 |
|
16 |
-
F1-Score: **
|
17 |
|
18 |
Predicts 4 tags:
|
19 |
|
@@ -37,10 +37,10 @@ from flair.data import Sentence
|
|
37 |
from flair.models import SequenceTagger
|
38 |
|
39 |
# load tagger
|
40 |
-
tagger = SequenceTagger.load("flair/ner-
|
41 |
|
42 |
# make example sentence
|
43 |
-
sentence = Sentence("George Washington
|
44 |
|
45 |
# predict NER tags
|
46 |
tagger.predict(sentence)
|
@@ -58,11 +58,11 @@ for entity in sentence.get_spans('ner'):
|
|
58 |
|
59 |
This yields the following output:
|
60 |
```
|
61 |
-
Span [1,2]: "George Washington" [− Labels: PER (0.
|
62 |
-
Span [
|
63 |
```
|
64 |
|
65 |
-
So, the entities "*George Washington*" (labeled as a **person**) and "*Washington*" (labeled as a **location**) are found in the sentence "*George Washington
|
66 |
|
67 |
|
68 |
---
|
@@ -73,11 +73,11 @@ The following Flair script was used to train this model:
|
|
73 |
|
74 |
```python
|
75 |
from flair.data import Corpus
|
76 |
-
from flair.datasets import
|
77 |
from flair.embeddings import WordEmbeddings, StackedEmbeddings, FlairEmbeddings
|
78 |
|
79 |
# 1. get the corpus
|
80 |
-
corpus: Corpus =
|
81 |
|
82 |
# 2. what tag do we want to predict?
|
83 |
tag_type = 'ner'
|
@@ -89,13 +89,13 @@ tag_dictionary = corpus.make_tag_dictionary(tag_type=tag_type)
|
|
89 |
embedding_types = [
|
90 |
|
91 |
# GloVe embeddings
|
92 |
-
WordEmbeddings('
|
93 |
|
94 |
# contextual string embeddings, forward
|
95 |
-
FlairEmbeddings('
|
96 |
|
97 |
# contextual string embeddings, backward
|
98 |
-
FlairEmbeddings('
|
99 |
]
|
100 |
|
101 |
# embedding stack consists of Flair and GloVe embeddings
|
@@ -115,7 +115,7 @@ from flair.trainers import ModelTrainer
|
|
115 |
trainer = ModelTrainer(tagger, corpus)
|
116 |
|
117 |
# 7. run training
|
118 |
-
trainer.train('resources/taggers/ner-
|
119 |
train_with_dev=True,
|
120 |
max_epochs=150)
|
121 |
```
|
|
|
9 |
inference: false
|
10 |
---
|
11 |
|
12 |
+
## French NER in Flair (default model)
|
13 |
|
14 |
+
This is the standard 4-class NER model for French that ships with [Flair](https://github.com/flairNLP/flair/).
|
15 |
|
16 |
+
F1-Score: **90,61** (WikiNER)
|
17 |
|
18 |
Predicts 4 tags:
|
19 |
|
|
|
37 |
from flair.models import SequenceTagger
|
38 |
|
39 |
# load tagger
|
40 |
+
tagger = SequenceTagger.load("flair/ner-french")
|
41 |
|
42 |
# make example sentence
|
43 |
+
sentence = Sentence("George Washington est allé à Washington")
|
44 |
|
45 |
# predict NER tags
|
46 |
tagger.predict(sentence)
|
|
|
58 |
|
59 |
This yields the following output:
|
60 |
```
|
61 |
+
Span [1,2]: "George Washington" [− Labels: PER (0.7394)]
|
62 |
+
Span [6]: "Washington" [− Labels: LOC (0.9161)]
|
63 |
```
|
64 |
|
65 |
+
So, the entities "*George Washington*" (labeled as a **person**) and "*Washington*" (labeled as a **location**) are found in the sentence "*George Washington est allé à Washington*".
|
66 |
|
67 |
|
68 |
---
|
|
|
73 |
|
74 |
```python
|
75 |
from flair.data import Corpus
|
76 |
+
from flair.datasets import WIKINER_FRENCH
|
77 |
from flair.embeddings import WordEmbeddings, StackedEmbeddings, FlairEmbeddings
|
78 |
|
79 |
# 1. get the corpus
|
80 |
+
corpus: Corpus = WIKINER_FRENCH()
|
81 |
|
82 |
# 2. what tag do we want to predict?
|
83 |
tag_type = 'ner'
|
|
|
89 |
embedding_types = [
|
90 |
|
91 |
# GloVe embeddings
|
92 |
+
WordEmbeddings('fr'),
|
93 |
|
94 |
# contextual string embeddings, forward
|
95 |
+
FlairEmbeddings('fr-forward'),
|
96 |
|
97 |
# contextual string embeddings, backward
|
98 |
+
FlairEmbeddings('fr-backward'),
|
99 |
]
|
100 |
|
101 |
# embedding stack consists of Flair and GloVe embeddings
|
|
|
115 |
trainer = ModelTrainer(tagger, corpus)
|
116 |
|
117 |
# 7. run training
|
118 |
+
trainer.train('resources/taggers/ner-french',
|
119 |
train_with_dev=True,
|
120 |
max_epochs=150)
|
121 |
```
|