|
1
|
|
|
"""Clases for supporting document corpora""" |
|
2
|
|
|
|
|
3
|
|
|
import glob |
|
4
|
|
|
import gzip |
|
5
|
|
|
import os.path |
|
6
|
|
|
import re |
|
7
|
|
|
from itertools import islice |
|
8
|
|
|
|
|
9
|
|
|
import annif.util |
|
10
|
|
|
|
|
11
|
|
|
from .subject import SubjectSet |
|
12
|
|
|
from .types import Document, DocumentCorpus |
|
13
|
|
|
|
|
14
|
|
|
logger = annif.logger |
|
15
|
|
|
|
|
16
|
|
|
|
|
17
|
|
|
class DocumentDirectory(DocumentCorpus): |
|
18
|
|
|
"""A directory of files as a full text document corpus""" |
|
19
|
|
|
|
|
20
|
|
|
def __init__(self, path, subject_index, language, require_subjects=False): |
|
21
|
|
|
self.path = path |
|
22
|
|
|
self.subject_index = subject_index |
|
23
|
|
|
self.language = language |
|
24
|
|
|
self.require_subjects = require_subjects |
|
25
|
|
|
|
|
26
|
|
|
def __iter__(self): |
|
27
|
|
|
"""Iterate through the directory, yielding tuples of (docfile, |
|
28
|
|
|
subjectfile) containing file paths. If there is no key file and |
|
29
|
|
|
require_subjects is False, the subjectfile will be returned as None.""" |
|
30
|
|
|
|
|
31
|
|
|
for filename in sorted(glob.glob(os.path.join(self.path, "*.txt"))): |
|
32
|
|
|
tsvfilename = re.sub(r"\.txt$", ".tsv", filename) |
|
33
|
|
|
if os.path.exists(tsvfilename): |
|
34
|
|
|
yield (filename, tsvfilename) |
|
35
|
|
|
continue |
|
36
|
|
|
keyfilename = re.sub(r"\.txt$", ".key", filename) |
|
37
|
|
|
if os.path.exists(keyfilename): |
|
38
|
|
|
yield (filename, keyfilename) |
|
39
|
|
|
continue |
|
40
|
|
|
if not self.require_subjects: |
|
41
|
|
|
yield (filename, None) |
|
42
|
|
|
|
|
43
|
|
|
@property |
|
44
|
|
|
def documents(self): |
|
45
|
|
|
for docfilename, keyfilename in self: |
|
46
|
|
|
with open(docfilename, errors="replace", encoding="utf-8-sig") as docfile: |
|
47
|
|
|
text = docfile.read() |
|
48
|
|
|
if keyfilename is None: |
|
49
|
|
|
yield Document(text=text, subject_set=None) |
|
50
|
|
|
continue |
|
51
|
|
|
with open(keyfilename, encoding="utf-8-sig") as keyfile: |
|
52
|
|
|
subjects = SubjectSet.from_string( |
|
53
|
|
|
keyfile.read(), self.subject_index, self.language |
|
54
|
|
|
) |
|
55
|
|
|
yield Document(text=text, subject_set=subjects) |
|
56
|
|
|
|
|
57
|
|
|
|
|
58
|
|
|
class DocumentFile(DocumentCorpus): |
|
59
|
|
|
"""A TSV file as a corpus of documents with subjects""" |
|
60
|
|
|
|
|
61
|
|
|
def __init__(self, path, subject_index): |
|
62
|
|
|
self.path = path |
|
63
|
|
|
self.subject_index = subject_index |
|
64
|
|
|
|
|
65
|
|
|
@property |
|
66
|
|
|
def documents(self): |
|
67
|
|
|
if self.path.endswith(".gz"): |
|
68
|
|
|
opener = gzip.open |
|
69
|
|
|
else: |
|
70
|
|
|
opener = open |
|
71
|
|
|
with opener(self.path, mode="rt", encoding="utf-8-sig") as tsvfile: |
|
72
|
|
|
for line in tsvfile: |
|
73
|
|
|
yield from self._parse_tsv_line(line) |
|
74
|
|
|
|
|
75
|
|
|
def _parse_tsv_line(self, line): |
|
76
|
|
|
if "\t" in line: |
|
77
|
|
|
text, uris = line.split("\t", maxsplit=1) |
|
78
|
|
|
subject_ids = { |
|
79
|
|
|
self.subject_index.by_uri(annif.util.cleanup_uri(uri)) |
|
80
|
|
|
for uri in uris.split() |
|
81
|
|
|
} |
|
82
|
|
|
yield Document(text=text, subject_set=SubjectSet(subject_ids)) |
|
83
|
|
|
else: |
|
84
|
|
|
logger.warning('Skipping invalid line (missing tab): "%s"', line.rstrip()) |
|
85
|
|
|
|
|
86
|
|
|
|
|
87
|
|
|
class DocumentList(DocumentCorpus): |
|
88
|
|
|
"""A document corpus based on a list of other iterable of Document |
|
89
|
|
|
objects""" |
|
90
|
|
|
|
|
91
|
|
|
def __init__(self, documents): |
|
92
|
|
|
self._documents = documents |
|
93
|
|
|
|
|
94
|
|
|
@property |
|
95
|
|
|
def documents(self): |
|
96
|
|
|
yield from self._documents |
|
97
|
|
|
|
|
98
|
|
|
|
|
99
|
|
|
class TransformingDocumentCorpus(DocumentCorpus): |
|
100
|
|
|
"""A document corpus that wraps another document corpus but transforms the |
|
101
|
|
|
documents using a given transform function""" |
|
102
|
|
|
|
|
103
|
|
|
def __init__(self, corpus, transform_fn): |
|
104
|
|
|
self._orig_corpus = corpus |
|
105
|
|
|
self._transform_fn = transform_fn |
|
106
|
|
|
|
|
107
|
|
|
@property |
|
108
|
|
|
def documents(self): |
|
109
|
|
|
for doc in self._orig_corpus.documents: |
|
110
|
|
|
yield Document( |
|
111
|
|
|
text=self._transform_fn(doc.text), subject_set=doc.subject_set |
|
112
|
|
|
) |
|
113
|
|
|
|
|
114
|
|
|
|
|
115
|
|
|
class LimitingDocumentCorpus(DocumentCorpus): |
|
116
|
|
|
"""A document corpus that wraps another document corpus but limits the |
|
117
|
|
|
number of documents to a given limit""" |
|
118
|
|
|
|
|
119
|
|
|
def __init__(self, corpus, docs_limit): |
|
120
|
|
|
self._orig_corpus = corpus |
|
121
|
|
|
self.docs_limit = docs_limit |
|
122
|
|
|
|
|
123
|
|
|
@property |
|
124
|
|
|
def documents(self): |
|
125
|
|
|
for doc in islice(self._orig_corpus.documents, self.docs_limit): |
|
126
|
|
|
yield doc |
|
127
|
|
|
|
|
128
|
|
|
|
|
129
|
|
|
class BatchingDocumentCorpus(DocumentCorpus): |
|
130
|
|
|
"""A document corpus that wraps another document corpus to allow iterating over the |
|
131
|
|
|
documents in batches of a given size; a batch is a list of documents.""" |
|
132
|
|
|
|
|
133
|
|
|
def __init__(self, corpus): |
|
134
|
|
|
self._orig_corpus = corpus |
|
135
|
|
|
|
|
136
|
|
|
@property |
|
137
|
|
|
def documents(self): |
|
138
|
|
|
yield from self._orig_corpus.documents |
|
139
|
|
|
|
|
140
|
|
|
def doc_batches(self, batch_size): |
|
141
|
|
|
it = iter(self.documents) |
|
142
|
|
|
while True: |
|
143
|
|
|
docs_batch = list(islice(it, batch_size)) |
|
144
|
|
|
if not docs_batch: |
|
145
|
|
|
return |
|
146
|
|
|
yield docs_batch |
|
147
|
|
|
|
NatLibFi /
Annif
