|
1
|
|
|
"""Maui backend that makes calls to a Maui Server instance using its API""" |
|
2
|
|
|
|
|
3
|
|
|
|
|
4
|
|
|
import time |
|
5
|
|
|
import os.path |
|
6
|
|
|
import json |
|
7
|
|
|
import requests |
|
8
|
|
|
import requests.exceptions |
|
9
|
|
|
from annif.exception import ConfigurationException |
|
10
|
|
|
from annif.exception import NotSupportedException |
|
11
|
|
|
from annif.exception import OperationFailedException |
|
12
|
|
|
from annif.suggestion import SubjectSuggestion, ListSuggestionResult |
|
13
|
|
|
from . import backend |
|
14
|
|
|
|
|
15
|
|
|
|
|
16
|
|
|
class MauiBackend(backend.AnnifBackend): |
|
17
|
|
|
name = "maui" |
|
18
|
|
|
|
|
19
|
|
|
TRAIN_FILE = 'maui-train.jsonl' |
|
20
|
|
|
|
|
21
|
|
|
@property |
|
22
|
|
|
def endpoint(self): |
|
23
|
|
|
try: |
|
24
|
|
|
return self.params['endpoint'] |
|
25
|
|
|
except KeyError: |
|
26
|
|
|
raise ConfigurationException( |
|
27
|
|
|
"endpoint must be set in project configuration", |
|
28
|
|
|
backend_id=self.backend_id) |
|
29
|
|
|
|
|
30
|
|
|
@property |
|
31
|
|
|
def tagger(self): |
|
32
|
|
|
try: |
|
33
|
|
|
return self.params['tagger'] |
|
34
|
|
|
except KeyError: |
|
35
|
|
|
raise ConfigurationException( |
|
36
|
|
|
"tagger must be set in project configuration", |
|
37
|
|
|
backend_id=self.backend_id) |
|
38
|
|
|
|
|
39
|
|
|
@property |
|
40
|
|
|
def tagger_url(self): |
|
41
|
|
|
return self.endpoint + self.tagger |
|
42
|
|
|
|
|
43
|
|
|
def _initialize_tagger(self): |
|
44
|
|
|
self.info("Initializing Maui Service tagger '{}'".format(self.tagger)) |
|
45
|
|
|
|
|
46
|
|
|
# try to delete the tagger in case it already exists |
|
47
|
|
|
resp = requests.delete(self.tagger_url) |
|
48
|
|
|
self.debug("Trying to delete tagger {} returned status code {}" |
|
49
|
|
|
.format(self.tagger, resp.status_code)) |
|
50
|
|
|
|
|
51
|
|
|
# create a new tagger |
|
52
|
|
|
data = {'id': self.tagger, 'lang': self.params['language']} |
|
53
|
|
|
try: |
|
54
|
|
|
resp = requests.post(self.endpoint, data=data) |
|
55
|
|
|
self.debug("Trying to create tagger {} returned status code {}" |
|
56
|
|
|
.format(self.tagger, resp.status_code)) |
|
57
|
|
|
resp.raise_for_status() |
|
58
|
|
|
except requests.exceptions.RequestException as err: |
|
59
|
|
|
raise OperationFailedException(err) |
|
60
|
|
|
|
|
61
|
|
|
def _upload_vocabulary(self, project): |
|
62
|
|
|
self.info("Uploading vocabulary") |
|
63
|
|
|
try: |
|
64
|
|
|
resp = requests.put(self.tagger_url + '/vocab', |
|
65
|
|
|
data=project.vocab.as_skos()) |
|
66
|
|
|
resp.raise_for_status() |
|
67
|
|
|
except requests.exceptions.RequestException as err: |
|
68
|
|
|
raise OperationFailedException(err) |
|
69
|
|
|
|
|
70
|
|
|
def _create_train_file(self, corpus): |
|
71
|
|
|
self.info("Creating train file") |
|
72
|
|
|
train_path = os.path.join(self.datadir, self.TRAIN_FILE) |
|
73
|
|
|
with open(train_path, 'w') as train_file: |
|
74
|
|
|
for doc in corpus.documents: |
|
75
|
|
|
doc_obj = {'content': doc.text, 'topics': list(doc.labels)} |
|
76
|
|
|
json_doc = json.dumps(doc_obj) |
|
77
|
|
|
print(json_doc, file=train_file) |
|
78
|
|
|
|
|
79
|
|
|
def _upload_train_file(self): |
|
80
|
|
|
self.info("Uploading training documents") |
|
81
|
|
|
train_path = os.path.join(self.datadir, self.TRAIN_FILE) |
|
82
|
|
|
with open(train_path, 'rb') as train_file: |
|
83
|
|
|
try: |
|
84
|
|
|
resp = requests.post(self.tagger_url + '/train', |
|
85
|
|
|
data=train_file) |
|
86
|
|
|
resp.raise_for_status() |
|
87
|
|
|
except requests.exceptions.RequestException as err: |
|
88
|
|
|
raise OperationFailedException(err) |
|
89
|
|
|
|
|
90
|
|
|
def _wait_for_train(self): |
|
91
|
|
|
self.info("Waiting for training to be completed...") |
|
92
|
|
|
while True: |
|
93
|
|
|
try: |
|
94
|
|
|
resp = requests.get(self.tagger_url + "/train") |
|
95
|
|
|
resp.raise_for_status() |
|
96
|
|
|
except requests.exceptions.RequestException as err: |
|
97
|
|
|
raise OperationFailedException(err) |
|
98
|
|
|
|
|
99
|
|
|
response = resp.json() |
|
100
|
|
|
if response['completed']: |
|
101
|
|
|
self.info("Training completed.") |
|
102
|
|
|
return |
|
103
|
|
|
time.sleep(1) |
|
104
|
|
|
|
|
105
|
|
|
def train(self, corpus, project): |
|
106
|
|
|
if corpus.is_empty(): |
|
107
|
|
|
raise NotSupportedException('training backend {} with no documents' |
|
108
|
|
|
.format(self.backend_id)) |
|
109
|
|
|
self._initialize_tagger() |
|
110
|
|
|
self._upload_vocabulary(project) |
|
111
|
|
|
self._create_train_file(corpus) |
|
112
|
|
|
self._upload_train_file() |
|
113
|
|
|
self._wait_for_train() |
|
114
|
|
|
|
|
115
|
|
|
def _suggest_request(self, text): |
|
116
|
|
|
data = {'text': text} |
|
117
|
|
|
|
|
118
|
|
|
try: |
|
119
|
|
|
resp = requests.post(self.tagger_url + '/suggest', data=data) |
|
120
|
|
|
resp.raise_for_status() |
|
121
|
|
|
except requests.exceptions.RequestException as err: |
|
122
|
|
|
self.warning("HTTP request failed: {}".format(err)) |
|
123
|
|
|
return None |
|
124
|
|
|
|
|
125
|
|
|
try: |
|
126
|
|
|
return resp.json() |
|
127
|
|
|
except ValueError as err: |
|
128
|
|
|
self.warning("JSON decode failed: {}".format(err)) |
|
129
|
|
|
return None |
|
130
|
|
|
|
|
131
|
|
|
def _response_to_result(self, response, project): |
|
132
|
|
|
try: |
|
133
|
|
|
return ListSuggestionResult( |
|
134
|
|
|
[SubjectSuggestion(uri=h['id'], |
|
135
|
|
|
label=h['label'], |
|
136
|
|
|
score=h['probability']) |
|
137
|
|
|
for h in response['topics'] |
|
138
|
|
|
if h['probability'] > 0.0], project.subjects) |
|
139
|
|
|
except (TypeError, ValueError) as err: |
|
140
|
|
|
self.warning("Problem interpreting JSON data: {}".format(err)) |
|
141
|
|
|
return ListSuggestionResult([], project.subjects) |
|
142
|
|
|
|
|
143
|
|
|
def _suggest(self, text, project, params): |
|
144
|
|
|
response = self._suggest_request(text) |
|
145
|
|
|
if response: |
|
146
|
|
|
return self._response_to_result(response, project) |
|
147
|
|
|
else: |
|
148
|
|
|
return ListSuggestionResult([], project.subjects) |
|
149
|
|
|
|
NatLibFi /
Annif
