|
1
|
|
|
"""Common functionality for backends.""" |
|
2
|
|
|
|
|
3
|
|
|
import abc |
|
4
|
|
|
import os.path |
|
5
|
|
|
from datetime import datetime, timezone |
|
6
|
|
|
from glob import glob |
|
7
|
|
|
from annif import logger |
|
8
|
|
|
from annif.corpus import TruncatingDocumentCorpus |
|
9
|
|
|
from annif.exception import ConfigurationException |
|
10
|
|
|
|
|
11
|
|
|
|
|
12
|
|
|
class AnnifBackend(metaclass=abc.ABCMeta): |
|
13
|
|
|
"""Base class for Annif backends that perform analysis. The |
|
14
|
|
|
non-implemented methods should be overridden in subclasses.""" |
|
15
|
|
|
|
|
16
|
|
|
name = None |
|
17
|
|
|
needs_subject_index = False |
|
18
|
|
|
|
|
19
|
|
|
DEFAULT_PARAMETERS = {'limit': 100, |
|
20
|
|
|
'input_limit': 0} |
|
21
|
|
|
|
|
22
|
|
|
def __init__(self, backend_id, config_params, project): |
|
23
|
|
|
"""Initialize backend with specific parameters. The |
|
24
|
|
|
parameters are a dict. Keys and values depend on the specific |
|
25
|
|
|
backend type.""" |
|
26
|
|
|
self.backend_id = backend_id |
|
27
|
|
|
self.config_params = config_params |
|
28
|
|
|
self.project = project |
|
29
|
|
|
self.datadir = project.datadir |
|
30
|
|
|
|
|
31
|
|
|
def default_params(self): |
|
32
|
|
|
return self.DEFAULT_PARAMETERS |
|
33
|
|
|
|
|
34
|
|
|
@property |
|
35
|
|
|
def params(self): |
|
36
|
|
|
params = {} |
|
37
|
|
|
params.update(self.default_params()) |
|
38
|
|
|
params.update(self.config_params) |
|
39
|
|
|
return params |
|
40
|
|
|
|
|
41
|
|
|
@property |
|
42
|
|
|
def is_trained(self): |
|
43
|
|
|
return bool(glob(os.path.join(self.datadir, '*'))) |
|
44
|
|
|
|
|
45
|
|
|
@property |
|
46
|
|
|
def modification_time(self): |
|
47
|
|
|
mtimes = [datetime.utcfromtimestamp(os.path.getmtime(p)) |
|
48
|
|
|
for p in glob(os.path.join(self.datadir, '*'))] |
|
49
|
|
|
most_recent = max(mtimes, default=None) |
|
50
|
|
|
if most_recent is None: |
|
51
|
|
|
return None |
|
52
|
|
|
return most_recent.replace(tzinfo=timezone.utc) |
|
53
|
|
|
|
|
54
|
|
|
def _get_backend_params(self, params): |
|
55
|
|
|
backend_params = dict(self.params) |
|
56
|
|
|
if params is not None: |
|
57
|
|
|
backend_params.update(params) |
|
58
|
|
|
return backend_params |
|
59
|
|
|
|
|
60
|
|
|
def _validate_input_limit(self, input_limit): |
|
61
|
|
|
input_limit = int(input_limit) |
|
62
|
|
|
if input_limit >= 0: |
|
63
|
|
|
return input_limit |
|
64
|
|
|
else: |
|
65
|
|
|
raise ConfigurationException( |
|
66
|
|
|
'input_limit can not be negative', backend_id=self.backend_id) |
|
67
|
|
|
|
|
68
|
|
|
def _train(self, corpus, params): |
|
69
|
|
|
"""This method can be overridden by backends. It implements |
|
70
|
|
|
the train functionality, with pre-processed parameters.""" |
|
71
|
|
|
pass # default is to do nothing, subclasses may override |
|
72
|
|
|
|
|
73
|
|
|
def train(self, corpus, params=None): |
|
74
|
|
|
"""Train the model on the given document or subject corpus.""" |
|
75
|
|
|
beparams = self._get_backend_params(params) |
|
76
|
|
|
input_limit = self._validate_input_limit(beparams['input_limit']) |
|
77
|
|
|
if input_limit != 0: |
|
78
|
|
|
corpus = TruncatingDocumentCorpus(corpus, input_limit) |
|
79
|
|
|
return self._train(corpus, params=beparams) |
|
80
|
|
|
|
|
81
|
|
|
def initialize(self): |
|
82
|
|
|
"""This method can be overridden by backends. It should cause the |
|
83
|
|
|
backend to pre-load all data it needs during operation.""" |
|
84
|
|
|
pass |
|
85
|
|
|
|
|
86
|
|
|
@abc.abstractmethod |
|
87
|
|
|
def _suggest(self, text, params): |
|
88
|
|
|
"""This method should implemented by backends. It implements |
|
89
|
|
|
the suggest functionality, with pre-processed parameters.""" |
|
90
|
|
|
pass # pragma: no cover |
|
91
|
|
|
|
|
92
|
|
|
def suggest(self, text, params=None): |
|
93
|
|
|
"""Suggest subjects for the input text and return a list of subjects |
|
94
|
|
|
represented as a list of SubjectSuggestion objects.""" |
|
95
|
|
|
beparams = self._get_backend_params(params) |
|
96
|
|
|
self.initialize() |
|
97
|
|
|
input_limit = self._validate_input_limit(beparams['input_limit']) |
|
98
|
|
|
if input_limit != 0: |
|
99
|
|
|
text = text[:input_limit] |
|
100
|
|
|
return self._suggest(text, params=beparams) |
|
101
|
|
|
|
|
102
|
|
|
def debug(self, message): |
|
103
|
|
|
"""Log a debug message from this backend""" |
|
104
|
|
|
logger.debug("Backend {}: {}".format(self.backend_id, message)) |
|
105
|
|
|
|
|
106
|
|
|
def info(self, message): |
|
107
|
|
|
"""Log an info message from this backend""" |
|
108
|
|
|
logger.info("Backend {}: {}".format(self.backend_id, message)) |
|
109
|
|
|
|
|
110
|
|
|
def warning(self, message): |
|
111
|
|
|
"""Log a warning message from this backend""" |
|
112
|
|
|
logger.warning("Backend {}: {}".format(self.backend_id, message)) |
|
113
|
|
|
|
|
114
|
|
|
|
|
115
|
|
|
class AnnifLearningBackend(AnnifBackend): |
|
116
|
|
|
"""Base class for Annif backends that can perform online learning""" |
|
117
|
|
|
|
|
118
|
|
|
@abc.abstractmethod |
|
119
|
|
|
def _learn(self, corpus, params): |
|
120
|
|
|
"""This method should implemented by backends. It implements the learn |
|
121
|
|
|
functionality, with pre-processed parameters.""" |
|
122
|
|
|
pass # pragma: no cover |
|
123
|
|
|
|
|
124
|
|
|
def learn(self, corpus, params=None): |
|
125
|
|
|
"""Further train the model on the given document or subject corpus.""" |
|
126
|
|
|
beparams = self._get_backend_params(params) |
|
127
|
|
|
input_limit = self._validate_input_limit(beparams['input_limit']) |
|
128
|
|
|
if input_limit != 0: |
|
129
|
|
|
corpus = TruncatingDocumentCorpus(corpus, input_limit) |
|
130
|
|
|
return self._learn(corpus, params=beparams) |
|
131
|
|
|
|
NatLibFi /
Annif
