|
1
|
|
|
"""Classes for supporting subject corpora expressed as directories or files""" |
|
2
|
|
|
|
|
3
|
|
|
import glob |
|
4
|
|
|
import os.path |
|
5
|
|
|
import annif.util |
|
6
|
|
|
import numpy as np |
|
7
|
|
|
from annif import logger |
|
8
|
|
|
from .types import Subject, SubjectCorpus |
|
9
|
|
|
from .convert import SubjectToDocumentCorpusMixin |
|
10
|
|
|
|
|
11
|
|
|
|
|
12
|
|
|
class SubjectDirectory(SubjectCorpus, SubjectToDocumentCorpusMixin): |
|
13
|
|
|
"""A subject corpus in the form of a directory with .txt files.""" |
|
14
|
|
|
|
|
15
|
|
|
def __init__(self, path): |
|
16
|
|
|
self.path = path |
|
17
|
|
|
self._filenames = sorted(glob.glob(os.path.join(path, '*.txt'))) |
|
18
|
|
|
|
|
19
|
|
|
@property |
|
20
|
|
|
def subjects(self): |
|
21
|
|
|
for filename in self._filenames: |
|
22
|
|
|
with open(filename) as subjfile: |
|
23
|
|
|
uri, label = subjfile.readline().strip().split(' ', 1) |
|
24
|
|
|
text = ' '.join(subjfile.readlines()) |
|
25
|
|
|
yield Subject(uri=uri, label=label, text=text) |
|
26
|
|
|
|
|
27
|
|
|
|
|
28
|
|
|
class SubjectFileTSV(SubjectCorpus, SubjectToDocumentCorpusMixin): |
|
29
|
|
|
"""A subject corpus stored in a TSV file.""" |
|
30
|
|
|
|
|
31
|
|
|
def __init__(self, path): |
|
32
|
|
|
self.path = path |
|
33
|
|
|
|
|
34
|
|
|
@property |
|
35
|
|
|
def subjects(self): |
|
36
|
|
|
with open(self.path) as subjfile: |
|
37
|
|
|
for line in subjfile: |
|
38
|
|
|
uri, label = line.strip().split(None, 1) |
|
39
|
|
|
clean_uri = annif.util.cleanup_uri(uri) |
|
40
|
|
|
yield Subject(uri=clean_uri, label=label, text=None) |
|
41
|
|
|
|
|
42
|
|
|
|
|
43
|
|
|
class SubjectIndex: |
|
44
|
|
|
"""An index that remembers the associations between integers subject IDs |
|
45
|
|
|
and their URIs and labels.""" |
|
46
|
|
|
|
|
47
|
|
|
def __init__(self, corpus): |
|
48
|
|
|
"""Initialize the subject index from a subject corpus.""" |
|
49
|
|
|
self._uris = [] |
|
50
|
|
|
self._labels = [] |
|
51
|
|
|
self._uri_idx = {} |
|
52
|
|
|
self._label_idx = {} |
|
53
|
|
|
for subject_id, subject in enumerate(corpus.subjects): |
|
54
|
|
|
self._uris.append(subject.uri) |
|
55
|
|
|
self._labels.append(subject.label) |
|
56
|
|
|
self._uri_idx[subject.uri] = subject_id |
|
57
|
|
|
self._label_idx[subject.label] = subject_id |
|
58
|
|
|
|
|
59
|
|
|
def __len__(self): |
|
60
|
|
|
return len(self._uris) |
|
61
|
|
|
|
|
62
|
|
|
def __getitem__(self, subject_id): |
|
63
|
|
|
return (self._uris[subject_id], self._labels[subject_id]) |
|
64
|
|
|
|
|
65
|
|
|
def by_uri(self, uri): |
|
66
|
|
|
"""return the subject index of a subject by its URI""" |
|
67
|
|
|
try: |
|
68
|
|
|
return self._uri_idx[uri] |
|
69
|
|
|
except KeyError: |
|
70
|
|
|
logger.warning('Unknown subject URI <%s>', uri) |
|
71
|
|
|
return None |
|
72
|
|
|
|
|
73
|
|
|
def by_label(self, label): |
|
74
|
|
|
"""return the subject index of a subject by its label""" |
|
75
|
|
|
try: |
|
76
|
|
|
return self._label_idx[label] |
|
77
|
|
|
except KeyError: |
|
78
|
|
|
logger.warning('Unknown subject label "%s"', label) |
|
79
|
|
|
return None |
|
80
|
|
|
|
|
81
|
|
|
def save(self, path): |
|
82
|
|
|
"""Save this subject index into a file.""" |
|
83
|
|
|
|
|
84
|
|
|
with open(path, 'w') as subjfile: |
|
85
|
|
|
for subject_id in range(len(self)): |
|
86
|
|
|
line = "<{}>\t{}".format( |
|
87
|
|
|
self._uris[subject_id], self._labels[subject_id]) |
|
88
|
|
|
print(line, file=subjfile) |
|
89
|
|
|
|
|
90
|
|
|
@classmethod |
|
91
|
|
|
def load(cls, path): |
|
92
|
|
|
"""Load a subject index from a TSV file and return it.""" |
|
93
|
|
|
|
|
94
|
|
|
corpus = SubjectFileTSV(path) |
|
95
|
|
|
return cls(corpus) |
|
96
|
|
|
|
|
97
|
|
|
|
|
98
|
|
|
class SubjectSet: |
|
99
|
|
|
"""Represents a set of subjects for a document.""" |
|
100
|
|
|
|
|
101
|
|
|
def __init__(self, subj_data=None): |
|
102
|
|
|
"""Create a SubjectSet and optionally initialize it from a tuple |
|
103
|
|
|
(URIs, labels)""" |
|
104
|
|
|
|
|
105
|
|
|
uris, labels = subj_data or ([], []) |
|
106
|
|
|
self.subject_uris = set(uris) |
|
107
|
|
|
self.subject_labels = set(labels) |
|
108
|
|
|
|
|
109
|
|
|
@classmethod |
|
110
|
|
|
def from_string(cls, subj_data): |
|
111
|
|
|
sset = cls() |
|
112
|
|
|
for line in subj_data.splitlines(): |
|
113
|
|
|
sset._parse_line(line) |
|
114
|
|
|
return sset |
|
115
|
|
|
|
|
116
|
|
|
def _parse_line(self, line): |
|
117
|
|
|
vals = line.split("\t") |
|
118
|
|
|
for val in vals: |
|
119
|
|
|
val = val.strip() |
|
120
|
|
|
if val == '': |
|
121
|
|
|
continue |
|
122
|
|
|
if val.startswith('<') and val.endswith('>'): # URI |
|
123
|
|
|
self.subject_uris.add(val[1:-1]) |
|
124
|
|
|
continue |
|
125
|
|
|
self.subject_labels.add(val) |
|
126
|
|
|
return |
|
127
|
|
|
|
|
128
|
|
|
def has_uris(self): |
|
129
|
|
|
"""returns True if the URIs for all subjects are known""" |
|
130
|
|
|
return len(self.subject_uris) >= len(self.subject_labels) |
|
131
|
|
|
|
|
132
|
|
|
def as_vector(self, subject_index): |
|
133
|
|
|
"""Return the hits as a one-dimensional NumPy array in sklearn |
|
134
|
|
|
multilabel indicator format, using a subject index as the source |
|
135
|
|
|
of subjects.""" |
|
136
|
|
|
|
|
137
|
|
|
vector = np.zeros(len(subject_index), dtype=np.int8) |
|
138
|
|
|
if self.has_uris(): |
|
139
|
|
|
for uri in self.subject_uris: |
|
140
|
|
|
subject_id = subject_index.by_uri(uri) |
|
141
|
|
|
if subject_id is not None: |
|
142
|
|
|
vector[subject_id] = 1 |
|
143
|
|
|
else: |
|
144
|
|
|
for label in self.subject_labels: |
|
145
|
|
|
subject_id = subject_index.by_label(label) |
|
146
|
|
|
if subject_id is not None: |
|
147
|
|
|
vector[subject_id] = 1 |
|
148
|
|
|
return vector |
|
149
|
|
|
|
NatLibFi /
Annif
