|
1
|
|
|
"""Annif backend mixins that can be used to implement features""" |
|
2
|
|
|
|
|
3
|
|
|
from __future__ import annotations |
|
4
|
|
|
|
|
5
|
|
|
import abc |
|
6
|
|
|
import os.path |
|
7
|
|
|
from typing import TYPE_CHECKING, Any |
|
8
|
|
|
|
|
9
|
|
|
import joblib |
|
10
|
|
|
import numpy as np |
|
11
|
|
|
from pecos.utils.featurization.text.vectorizers import Vectorizer |
|
12
|
|
|
from sklearn.feature_extraction.text import TfidfVectorizer |
|
13
|
|
|
|
|
14
|
|
|
import annif.util |
|
15
|
|
|
from annif.exception import NotInitializedException |
|
16
|
|
|
|
|
17
|
|
|
if TYPE_CHECKING: |
|
18
|
|
|
from collections.abc import Iterable |
|
19
|
|
|
|
|
20
|
|
|
from scipy.sparse._csr import csr_matrix |
|
21
|
|
|
|
|
22
|
|
|
from annif.corpus import Document |
|
23
|
|
|
from annif.suggestion import SubjectSuggestion |
|
24
|
|
|
|
|
25
|
|
|
|
|
26
|
|
|
class ChunkingBackend(metaclass=abc.ABCMeta): |
|
27
|
|
|
"""Annif backend mixin that implements chunking of input""" |
|
28
|
|
|
|
|
29
|
|
|
DEFAULT_PARAMETERS = {"chunksize": 1} |
|
30
|
|
|
|
|
31
|
|
|
def default_params(self) -> dict[str, Any]: |
|
32
|
|
|
return self.DEFAULT_PARAMETERS |
|
33
|
|
|
|
|
34
|
|
|
@abc.abstractmethod |
|
35
|
|
|
def _suggest_chunks( |
|
36
|
|
|
self, chunktexts: list[str], params: dict[str, Any] |
|
37
|
|
|
) -> list[SubjectSuggestion]: |
|
38
|
|
|
"""Suggest subjects for the chunked text; should be implemented by |
|
39
|
|
|
the subclass inheriting this mixin""" |
|
40
|
|
|
|
|
41
|
|
|
pass # pragma: no cover |
|
42
|
|
|
|
|
43
|
|
|
def _suggest( |
|
44
|
|
|
self, doc: Document, params: dict[str, Any] |
|
45
|
|
|
) -> list[SubjectSuggestion]: |
|
46
|
|
|
self.debug( |
|
47
|
|
|
'Suggesting subjects for text "{}..." (len={})'.format( |
|
48
|
|
|
doc.text[:20], len(doc.text) |
|
49
|
|
|
) |
|
50
|
|
|
) |
|
51
|
|
|
sentences = self.project.analyzer.tokenize_sentences(doc.text) |
|
52
|
|
|
self.debug("Found {} sentences".format(len(sentences))) |
|
53
|
|
|
chunksize = int(params["chunksize"]) |
|
54
|
|
|
chunktexts = [] |
|
55
|
|
|
for i in range(0, len(sentences), chunksize): |
|
56
|
|
|
chunktexts.append(" ".join(sentences[i : i + chunksize])) |
|
57
|
|
|
self.debug("Split sentences into {} chunks".format(len(chunktexts))) |
|
58
|
|
|
if len(chunktexts) == 0: # no input, empty result |
|
59
|
|
|
return [] |
|
60
|
|
|
return self._suggest_chunks(chunktexts, params) |
|
61
|
|
|
|
|
62
|
|
|
class TfidfVectorizerMixin: |
|
63
|
|
|
"""Annif backend mixin that implements TfidfVectorizer functionality""" |
|
64
|
|
|
|
|
65
|
|
|
VECTORIZER_FILE = "vectorizer" |
|
66
|
|
|
|
|
67
|
|
|
vectorizer = None |
|
68
|
|
|
|
|
69
|
|
|
def initialize_vectorizer(self) -> None: |
|
70
|
|
|
if self.vectorizer is None: |
|
71
|
|
|
path = os.path.join(self.datadir, self.VECTORIZER_FILE) |
|
72
|
|
|
if os.path.exists(path): |
|
73
|
|
|
self.debug("loading vectorizer from {}".format(path)) |
|
74
|
|
|
self.vectorizer = joblib.load(path) |
|
75
|
|
|
else: |
|
76
|
|
|
raise NotInitializedException( |
|
77
|
|
|
"vectorizer file '{}' not found".format(path), |
|
78
|
|
|
backend_id=self.backend_id, |
|
79
|
|
|
) |
|
80
|
|
|
|
|
81
|
|
|
def create_vectorizer( |
|
82
|
|
|
self, input: Iterable[str], params: dict[str, Any] = None |
|
83
|
|
|
) -> csr_matrix: |
|
84
|
|
|
self.info("creating vectorizer") |
|
85
|
|
|
if params is None: |
|
86
|
|
|
params = {} |
|
87
|
|
|
# avoid UserWarning when overriding tokenizer |
|
88
|
|
|
if "tokenizer" in params: |
|
89
|
|
|
params["token_pattern"] = None |
|
90
|
|
|
self.vectorizer = TfidfVectorizer(**params) |
|
91
|
|
|
veccorpus = self.vectorizer.fit_transform(input) |
|
92
|
|
|
annif.util.atomic_save( |
|
93
|
|
|
self.vectorizer, self.datadir, self.VECTORIZER_FILE, method=joblib.dump |
|
94
|
|
|
) |
|
95
|
|
|
return veccorpus |
|
96
|
|
|
|
|
97
|
|
|
class PecosTfidfVectorizerMixin: |
|
98
|
|
|
"""Annif backend mixin that implements TfidfVectorizer functionality from Pecos""" |
|
99
|
|
|
|
|
100
|
|
|
VECTORIZER_FILE = "vectorizer" |
|
101
|
|
|
|
|
102
|
|
|
vectorizer = None |
|
103
|
|
|
|
|
104
|
|
|
def initialize_vectorizer(self) -> None: |
|
105
|
|
|
if self.vectorizer is None: |
|
106
|
|
|
path = os.path.join(self.datadir, self.VECTORIZER_FILE) |
|
107
|
|
|
if os.path.exists(path): |
|
108
|
|
|
self.debug("loading vectorizer from {}".format(path)) |
|
109
|
|
|
|
|
110
|
|
|
self.vectorizer = Vectorizer.load(path) |
|
111
|
|
|
else: |
|
112
|
|
|
raise NotInitializedException( |
|
113
|
|
|
"vectorizer file '{}' not found".format(path), |
|
114
|
|
|
backend_id=self.backend_id, |
|
115
|
|
|
) |
|
116
|
|
|
|
|
117
|
|
|
def vectorizer_dict(self, params: dict[str, Any]) -> dict[str, Any]: |
|
118
|
|
|
"""Create a vectorizer configuration dictionary from the given parameters.""" |
|
119
|
|
|
|
|
120
|
|
|
config = { |
|
121
|
|
|
"base_vect_configs": [ |
|
122
|
|
|
{ |
|
123
|
|
|
"ngram_range": params.get("ngram_range", [1, 1]), |
|
124
|
|
|
"max_df_ratio": 0.98, |
|
125
|
|
|
"analyzer": "word", |
|
126
|
|
|
"min_df_cnt": params.get("min_df", 1), |
|
127
|
|
|
} |
|
128
|
|
|
] |
|
129
|
|
|
} |
|
130
|
|
|
return {"type": "tfidf", "kwargs": {**config}} |
|
131
|
|
|
|
|
132
|
|
|
|
|
133
|
|
|
def create_vectorizer( |
|
134
|
|
|
self, input: Iterable[str], params: dict[str, Any] = None |
|
135
|
|
|
) -> csr_matrix: |
|
136
|
|
|
|
|
137
|
|
|
self.info("creating Pecos vectorizer") |
|
138
|
|
|
if params is None: |
|
139
|
|
|
params = {} |
|
140
|
|
|
data = list(input) |
|
141
|
|
|
vectorizer_config = self.vectorizer_dict(params) |
|
142
|
|
|
self.vectorizer = Vectorizer.train(data, vectorizer_config, np.float32) |
|
143
|
|
|
self.vectorizer.save(os.path.join(self.datadir, self.VECTORIZER_FILE)) |
|
144
|
|
|
veccorpus = self.vectorizer.predict( |
|
145
|
|
|
data, |
|
146
|
|
|
threads=params.get("threads", -1) |
|
147
|
|
|
) |
|
148
|
|
|
|
|
149
|
|
|
return veccorpus |
|
150
|
|
|
|
NatLibFi /
Annif
