|
1
|
|
|
"""Common functionality for backends.""" |
|
2
|
|
|
|
|
3
|
|
|
from __future__ import annotations |
|
4
|
|
|
|
|
5
|
|
|
import abc |
|
6
|
|
|
import os.path |
|
7
|
|
|
from datetime import datetime, timezone |
|
8
|
|
|
from glob import glob |
|
9
|
|
|
from typing import TYPE_CHECKING, Any |
|
10
|
|
|
|
|
11
|
|
|
from annif import logger |
|
12
|
|
|
from annif.suggestion import SuggestionBatch |
|
13
|
|
|
|
|
14
|
|
|
if TYPE_CHECKING: |
|
15
|
|
|
from configparser import SectionProxy |
|
16
|
|
|
|
|
17
|
|
|
from annif.corpus.document import Document, DocumentCorpus |
|
18
|
|
|
from annif.project import AnnifProject |
|
19
|
|
|
|
|
20
|
|
|
|
|
21
|
|
|
class AnnifBackend(metaclass=abc.ABCMeta): |
|
22
|
|
|
"""Base class for Annif backends that perform analysis. The |
|
23
|
|
|
non-implemented methods should be overridden in subclasses.""" |
|
24
|
|
|
|
|
25
|
|
|
name = None |
|
26
|
|
|
|
|
27
|
|
|
DEFAULT_PARAMETERS = {"limit": 100} |
|
28
|
|
|
|
|
29
|
|
|
def __init__( |
|
30
|
|
|
self, |
|
31
|
|
|
backend_id: str, |
|
32
|
|
|
config_params: dict[str, Any] | SectionProxy, |
|
33
|
|
|
project: AnnifProject, |
|
34
|
|
|
) -> None: |
|
35
|
|
|
"""Initialize backend with specific parameters. The |
|
36
|
|
|
parameters are a dict. Keys and values depend on the specific |
|
37
|
|
|
backend type.""" |
|
38
|
|
|
self.backend_id = backend_id |
|
39
|
|
|
self.config_params = config_params |
|
40
|
|
|
self.project = project |
|
41
|
|
|
self.datadir = project.datadir |
|
42
|
|
|
|
|
43
|
|
|
def default_params(self) -> dict[str, Any]: |
|
44
|
|
|
params = AnnifBackend.DEFAULT_PARAMETERS.copy() |
|
45
|
|
|
params.update(self.DEFAULT_PARAMETERS) # Optional backend specific parameters |
|
46
|
|
|
return params |
|
47
|
|
|
|
|
48
|
|
|
@property |
|
49
|
|
|
def params(self) -> dict[str, Any]: |
|
50
|
|
|
params = {} |
|
51
|
|
|
params.update(self.default_params()) |
|
52
|
|
|
params.update(self.config_params) |
|
53
|
|
|
return params |
|
54
|
|
|
|
|
55
|
|
|
@property |
|
56
|
|
|
def _model_file_paths(self) -> list: |
|
57
|
|
|
all_paths = glob(os.path.join(self.datadir, "**"), recursive=True) |
|
58
|
|
|
file_paths = [p for p in all_paths if os.path.isfile(p)] |
|
59
|
|
|
ignore_patterns = ("*-train*", "tmp-*", "vectorizer") |
|
60
|
|
|
ignore_paths = [ |
|
61
|
|
|
path |
|
62
|
|
|
for igp in ignore_patterns |
|
63
|
|
|
for path in glob(os.path.join(self.datadir, igp)) |
|
64
|
|
|
] |
|
65
|
|
|
return list(set(file_paths) - set(ignore_paths)) |
|
66
|
|
|
|
|
67
|
|
|
@property |
|
68
|
|
|
def is_trained(self) -> bool: |
|
69
|
|
|
return bool(self._model_file_paths) |
|
70
|
|
|
|
|
71
|
|
|
@property |
|
72
|
|
|
def modification_time(self) -> datetime | None: |
|
73
|
|
|
mtimes = [ |
|
74
|
|
|
datetime.fromtimestamp(os.path.getmtime(p), tz=timezone.utc) |
|
75
|
|
|
for p in self._model_file_paths |
|
76
|
|
|
] |
|
77
|
|
|
most_recent = max(mtimes, default=None) |
|
78
|
|
|
return most_recent |
|
79
|
|
|
|
|
80
|
|
|
def _get_backend_params( |
|
81
|
|
|
self, |
|
82
|
|
|
params: dict[str, Any] | None, |
|
83
|
|
|
) -> dict[str, Any]: |
|
84
|
|
|
backend_params = dict(self.params) |
|
85
|
|
|
if params is not None: |
|
86
|
|
|
backend_params.update(params) |
|
87
|
|
|
return backend_params |
|
88
|
|
|
|
|
89
|
|
|
def _train( |
|
90
|
|
|
self, |
|
91
|
|
|
corpus: DocumentCorpus, |
|
92
|
|
|
params: dict[str, Any], |
|
93
|
|
|
jobs: int = 0, |
|
94
|
|
|
) -> None: |
|
95
|
|
|
"""This method can be overridden by backends. It implements |
|
96
|
|
|
the train functionality, with pre-processed parameters.""" |
|
97
|
|
|
pass # default is to do nothing, subclasses may override |
|
98
|
|
|
|
|
99
|
|
|
def train( |
|
100
|
|
|
self, |
|
101
|
|
|
corpus: DocumentCorpus, |
|
102
|
|
|
params: dict[str, Any] | None = None, |
|
103
|
|
|
jobs: int = 0, |
|
104
|
|
|
) -> None: |
|
105
|
|
|
"""Train the model on the given document or subject corpus.""" |
|
106
|
|
|
beparams = self._get_backend_params(params) |
|
107
|
|
|
return self._train(corpus, params=beparams, jobs=jobs) |
|
108
|
|
|
|
|
109
|
|
|
def initialize(self, parallel: bool = False) -> None: |
|
110
|
|
|
"""This method can be overridden by backends. It should cause the |
|
111
|
|
|
backend to pre-load all data it needs during operation. |
|
112
|
|
|
If parallel is True, the backend should expect to be used for |
|
113
|
|
|
parallel operation.""" |
|
114
|
|
|
pass |
|
115
|
|
|
|
|
116
|
|
|
def _suggest(self, doc: Document, params: dict[str, Any]): |
|
117
|
|
|
"""Either this method or _suggest_batch should be implemented by by |
|
118
|
|
|
backends. It implements the suggest functionality for a single |
|
119
|
|
|
document, with pre-processed parameters.""" |
|
120
|
|
|
pass # pragma: no cover |
|
121
|
|
|
|
|
122
|
|
|
def _suggest_batch( |
|
123
|
|
|
self, documents: list[Document], params: dict[str, Any] |
|
124
|
|
|
) -> SuggestionBatch: |
|
125
|
|
|
"""This method can be implemented by backends to use batching of documents in |
|
126
|
|
|
their operations. This default implementation uses the regular suggest |
|
127
|
|
|
functionality.""" |
|
128
|
|
|
return SuggestionBatch.from_sequence( |
|
129
|
|
|
[self._suggest(doc, params) for doc in documents], |
|
130
|
|
|
self.project.subjects, |
|
131
|
|
|
limit=int(params.get("limit")), |
|
132
|
|
|
) |
|
133
|
|
|
|
|
134
|
|
|
def suggest( |
|
135
|
|
|
self, |
|
136
|
|
|
documents: list[Document], |
|
137
|
|
|
params: dict[str, Any] | None = None, |
|
138
|
|
|
) -> SuggestionBatch: |
|
139
|
|
|
"""Suggest subjects for the input documents and return a list of subject sets |
|
140
|
|
|
represented as a list of SubjectSuggestion objects.""" |
|
141
|
|
|
beparams = self._get_backend_params(params) |
|
142
|
|
|
self.initialize() |
|
143
|
|
|
return self._suggest_batch(documents, params=beparams) |
|
144
|
|
|
|
|
145
|
|
|
def debug(self, message: str) -> None: |
|
146
|
|
|
"""Log a debug message from this backend""" |
|
147
|
|
|
logger.debug("Backend {}: {}".format(self.backend_id, message)) |
|
148
|
|
|
|
|
149
|
|
|
def info(self, message: str) -> None: |
|
150
|
|
|
"""Log an info message from this backend""" |
|
151
|
|
|
logger.info("Backend {}: {}".format(self.backend_id, message)) |
|
152
|
|
|
|
|
153
|
|
|
def warning(self, message: str) -> None: |
|
154
|
|
|
"""Log a warning message from this backend""" |
|
155
|
|
|
logger.warning("Backend {}: {}".format(self.backend_id, message)) |
|
156
|
|
|
|
|
157
|
|
|
|
|
158
|
|
|
class AnnifLearningBackend(AnnifBackend): |
|
159
|
|
|
"""Base class for Annif backends that can perform online learning""" |
|
160
|
|
|
|
|
161
|
|
|
@abc.abstractmethod |
|
162
|
|
|
def _learn(self, corpus, params): |
|
163
|
|
|
"""This method should implemented by backends. It implements the learn |
|
164
|
|
|
functionality, with pre-processed parameters.""" |
|
165
|
|
|
pass # pragma: no cover |
|
166
|
|
|
|
|
167
|
|
|
def learn( |
|
168
|
|
|
self, |
|
169
|
|
|
corpus: DocumentCorpus, |
|
170
|
|
|
params: dict[str, Any] | None = None, |
|
171
|
|
|
) -> None: |
|
172
|
|
|
"""Further train the model on the given document or subject corpus.""" |
|
173
|
|
|
beparams = self._get_backend_params(params) |
|
174
|
|
|
return self._learn(corpus, params=beparams) |
|
175
|
|
|
|
NatLibFi /
Annif
