|
1
|
|
|
"""Annif backend mixins that can be used to implement features""" |
|
2
|
|
|
|
|
3
|
|
|
|
|
4
|
|
|
import abc |
|
5
|
|
|
import os.path |
|
6
|
|
|
import joblib |
|
7
|
|
|
from sklearn.feature_extraction.text import TfidfVectorizer |
|
8
|
|
|
import annif.util |
|
9
|
|
|
from annif.exception import NotInitializedException |
|
10
|
|
|
from annif.suggestion import ListSuggestionResult |
|
11
|
|
|
|
|
12
|
|
|
|
|
13
|
|
|
class ChunkingBackend(metaclass=abc.ABCMeta): |
|
14
|
|
|
"""Annif backend mixin that implements chunking of input""" |
|
15
|
|
|
|
|
16
|
|
|
DEFAULT_PARAMS = {'chunksize': 1} |
|
17
|
|
|
|
|
18
|
|
|
def default_params(self): |
|
19
|
|
|
return self.DEFAULT_PARAMS |
|
20
|
|
|
|
|
21
|
|
|
@abc.abstractmethod |
|
22
|
|
|
def _suggest_chunks(self, chunktexts): |
|
23
|
|
|
"""Suggest subjects for the chunked text; should be implemented by |
|
24
|
|
|
the subclass inheriting this mixin""" |
|
25
|
|
|
|
|
26
|
|
|
pass # pragma: no cover |
|
27
|
|
|
|
|
28
|
|
|
def _suggest(self, text, params): |
|
29
|
|
|
self.debug('Suggesting subjects for text "{}..." (len={})'.format( |
|
30
|
|
|
text[:20], len(text))) |
|
31
|
|
|
sentences = self.project.analyzer.tokenize_sentences(text) |
|
32
|
|
|
self.debug('Found {} sentences'.format(len(sentences))) |
|
33
|
|
|
chunksize = int(params['chunksize']) |
|
34
|
|
|
chunktexts = [] |
|
35
|
|
|
for i in range(0, len(sentences), chunksize): |
|
36
|
|
|
chunktexts.append(' '.join(sentences[i:i + chunksize])) |
|
37
|
|
|
self.debug('Split sentences into {} chunks'.format(len(chunktexts))) |
|
38
|
|
|
if len(chunktexts) == 0: # no input, empty result |
|
39
|
|
|
return ListSuggestionResult( |
|
40
|
|
|
hits=[], subject_index=self.project.subjects) |
|
41
|
|
|
return self._suggest_chunks(chunktexts) |
|
42
|
|
|
|
|
43
|
|
|
|
|
44
|
|
|
class TfidfVectorizerMixin: |
|
45
|
|
|
"""Annif backend mixin that implements TfidfVectorizer functionality""" |
|
46
|
|
|
|
|
47
|
|
|
VECTORIZER_FILE = 'vectorizer' |
|
48
|
|
|
|
|
49
|
|
|
vectorizer = None |
|
50
|
|
|
|
|
51
|
|
|
def initialize_vectorizer(self): |
|
52
|
|
|
if self.vectorizer is None: |
|
53
|
|
|
path = os.path.join(self.datadir, self.VECTORIZER_FILE) |
|
54
|
|
|
if os.path.exists(path): |
|
55
|
|
|
self.debug('loading vectorizer from {}'.format(path)) |
|
56
|
|
|
self.vectorizer = joblib.load(path) |
|
57
|
|
|
else: |
|
58
|
|
|
raise NotInitializedException( |
|
59
|
|
|
"vectorizer file '{}' not found".format(path), |
|
60
|
|
|
backend_id=self.backend_id) |
|
61
|
|
|
|
|
62
|
|
|
def create_vectorizer(self, input, params={}): |
|
63
|
|
|
self.info('creating vectorizer') |
|
64
|
|
|
self.vectorizer = TfidfVectorizer(**params) |
|
65
|
|
|
veccorpus = self.vectorizer.fit_transform(input) |
|
66
|
|
|
annif.util.atomic_save( |
|
67
|
|
|
self.vectorizer, |
|
68
|
|
|
self.datadir, |
|
69
|
|
|
self.VECTORIZER_FILE, |
|
70
|
|
|
method=joblib.dump) |
|
71
|
|
|
return veccorpus |
|
72
|
|
|
|
NatLibFi /
Annif
