|
1
|
|
|
"""Definitions for command-line (Click) commands for invoking Annif |
|
2
|
|
|
operations and printing the results to console.""" |
|
3
|
|
|
|
|
4
|
|
|
|
|
5
|
|
|
import collections |
|
6
|
|
|
import os.path |
|
7
|
|
|
import re |
|
8
|
|
|
import sys |
|
9
|
|
|
import json |
|
10
|
|
|
import click |
|
11
|
|
|
import click_log |
|
12
|
|
|
from flask import current_app |
|
13
|
|
|
from flask.cli import FlaskGroup, ScriptInfo |
|
14
|
|
|
import annif |
|
15
|
|
|
import annif.corpus |
|
16
|
|
|
import annif.parallel |
|
17
|
|
|
import annif.project |
|
18
|
|
|
import annif.registry |
|
19
|
|
|
from annif.project import Access |
|
20
|
|
|
from annif.suggestion import SuggestionFilter, ListSuggestionResult |
|
21
|
|
|
from annif.exception import ConfigurationException, NotSupportedException |
|
22
|
|
|
from annif.exception import NotInitializedException |
|
23
|
|
|
from annif.util import metric_code |
|
24
|
|
|
|
|
25
|
|
|
logger = annif.logger |
|
26
|
|
|
click_log.basic_config(logger) |
|
27
|
|
|
|
|
28
|
|
|
cli = FlaskGroup(create_app=annif.create_app, add_version_option=False) |
|
29
|
|
|
cli = click.version_option(message='%(version)s')(cli) |
|
30
|
|
|
|
|
31
|
|
|
|
|
32
|
|
|
def get_project(project_id): |
|
33
|
|
|
""" |
|
34
|
|
|
Helper function to get a project by ID and bail out if it doesn't exist""" |
|
35
|
|
|
try: |
|
36
|
|
|
return annif.registry.get_project(project_id, |
|
37
|
|
|
min_access=Access.private) |
|
38
|
|
|
except ValueError: |
|
39
|
|
|
click.echo( |
|
40
|
|
|
"No projects found with id \'{0}\'.".format(project_id), |
|
41
|
|
|
err=True) |
|
42
|
|
|
sys.exit(1) |
|
43
|
|
|
|
|
44
|
|
|
|
|
45
|
|
|
def get_vocab(vocab_id): |
|
46
|
|
|
""" |
|
47
|
|
|
Helper function to get a vocabulary by ID and bail out if it doesn't |
|
48
|
|
|
exist""" |
|
49
|
|
|
try: |
|
50
|
|
|
return annif.registry.get_vocab(vocab_id, |
|
51
|
|
|
min_access=Access.private) |
|
52
|
|
|
except ValueError: |
|
53
|
|
|
click.echo( |
|
54
|
|
|
f"No vocabularies found with the id '{vocab_id}'.", |
|
55
|
|
|
err=True) |
|
56
|
|
|
sys.exit(1) |
|
57
|
|
|
|
|
58
|
|
|
|
|
59
|
|
|
def open_documents(paths, subject_index, vocab_lang, docs_limit): |
|
60
|
|
|
"""Helper function to open a document corpus from a list of pathnames, |
|
61
|
|
|
each of which is either a TSV file or a directory of TXT files. For |
|
62
|
|
|
directories with subjects in TSV files, the given vocabulary language |
|
63
|
|
|
will be used to convert subject labels into URIs. The corpus will be |
|
64
|
|
|
returned as an instance of DocumentCorpus or LimitingDocumentCorpus.""" |
|
65
|
|
|
|
|
66
|
|
|
def open_doc_path(path, subject_index): |
|
67
|
|
|
"""open a single path and return it as a DocumentCorpus""" |
|
68
|
|
|
if os.path.isdir(path): |
|
69
|
|
|
return annif.corpus.DocumentDirectory(path, subject_index, |
|
70
|
|
|
vocab_lang, |
|
71
|
|
|
require_subjects=True) |
|
72
|
|
|
return annif.corpus.DocumentFile(path, subject_index) |
|
73
|
|
|
|
|
74
|
|
|
if len(paths) == 0: |
|
75
|
|
|
logger.warning('Reading empty file') |
|
76
|
|
|
docs = open_doc_path(os.path.devnull, subject_index) |
|
77
|
|
|
elif len(paths) == 1: |
|
78
|
|
|
docs = open_doc_path(paths[0], subject_index) |
|
79
|
|
|
else: |
|
80
|
|
|
corpora = [open_doc_path(path, subject_index) for path in paths] |
|
81
|
|
|
docs = annif.corpus.CombinedCorpus(corpora) |
|
82
|
|
|
if docs_limit is not None: |
|
83
|
|
|
docs = annif.corpus.LimitingDocumentCorpus(docs, docs_limit) |
|
84
|
|
|
return docs |
|
85
|
|
|
|
|
86
|
|
|
|
|
87
|
|
|
def parse_backend_params(backend_param, project): |
|
88
|
|
|
"""Parse a list of backend parameters given with the --backend-param |
|
89
|
|
|
option into a nested dict structure""" |
|
90
|
|
|
backend_params = collections.defaultdict(dict) |
|
91
|
|
|
for beparam in backend_param: |
|
92
|
|
|
backend, param = beparam.split('.', 1) |
|
93
|
|
|
key, val = param.split('=', 1) |
|
94
|
|
|
validate_backend_params(backend, beparam, project) |
|
95
|
|
|
backend_params[backend][key] = val |
|
96
|
|
|
return backend_params |
|
97
|
|
|
|
|
98
|
|
|
|
|
99
|
|
|
def validate_backend_params(backend, beparam, project): |
|
100
|
|
|
if backend != project.config['backend']: |
|
101
|
|
|
raise ConfigurationException( |
|
102
|
|
|
'The backend {} in CLI option "-b {}" not matching the project' |
|
103
|
|
|
' backend {}.' |
|
104
|
|
|
.format(backend, beparam, project.config['backend'])) |
|
105
|
|
|
|
|
106
|
|
|
|
|
107
|
|
|
BATCH_MAX_LIMIT = 15 |
|
108
|
|
|
|
|
109
|
|
|
|
|
110
|
|
|
def generate_filter_batches(subjects): |
|
111
|
|
|
import annif.eval |
|
112
|
|
|
filter_batches = collections.OrderedDict() |
|
113
|
|
|
for limit in range(1, BATCH_MAX_LIMIT + 1): |
|
114
|
|
|
for threshold in [i * 0.05 for i in range(20)]: |
|
115
|
|
|
hit_filter = SuggestionFilter(subjects, limit, threshold) |
|
116
|
|
|
batch = annif.eval.EvaluationBatch(subjects) |
|
117
|
|
|
filter_batches[(limit, threshold)] = (hit_filter, batch) |
|
118
|
|
|
return filter_batches |
|
119
|
|
|
|
|
120
|
|
|
|
|
121
|
|
|
def set_project_config_file_path(ctx, param, value): |
|
122
|
|
|
"""Override the default path or the path given in env by CLI option""" |
|
123
|
|
|
with ctx.ensure_object(ScriptInfo).load_app().app_context(): |
|
124
|
|
|
if value: |
|
125
|
|
|
current_app.config['PROJECTS_CONFIG_PATH'] = value |
|
126
|
|
|
|
|
127
|
|
|
|
|
128
|
|
|
def common_options(f): |
|
129
|
|
|
"""Decorator to add common options for all CLI commands""" |
|
130
|
|
|
f = click.option( |
|
131
|
|
|
'-p', '--projects', |
|
132
|
|
|
help='Set path to project configuration file or directory', |
|
133
|
|
|
type=click.Path(dir_okay=True, exists=True), |
|
134
|
|
|
callback=set_project_config_file_path, expose_value=False, |
|
135
|
|
|
is_eager=True)(f) |
|
136
|
|
|
return click_log.simple_verbosity_option(logger)(f) |
|
137
|
|
|
|
|
138
|
|
|
|
|
139
|
|
|
def backend_param_option(f): |
|
140
|
|
|
"""Decorator to add an option for CLI commands to override BE parameters""" |
|
141
|
|
|
return click.option( |
|
142
|
|
|
'--backend-param', '-b', multiple=True, |
|
143
|
|
|
help='Override backend parameter of the config file. ' + |
|
144
|
|
|
'Syntax: `-b <backend>.<parameter>=<value>`.')(f) |
|
145
|
|
|
|
|
146
|
|
|
|
|
147
|
|
|
@cli.command('list-projects') |
|
148
|
|
|
@common_options |
|
149
|
|
|
@click_log.simple_verbosity_option(logger, default='ERROR') |
|
150
|
|
|
def run_list_projects(): |
|
151
|
|
|
""" |
|
152
|
|
|
List available projects. |
|
153
|
|
|
\f |
|
154
|
|
|
Show a list of currently defined projects. Projects are defined in a |
|
155
|
|
|
configuration file, normally called ``projects.cfg``. See `Project |
|
156
|
|
|
configuration |
|
157
|
|
|
<https://github.com/NatLibFi/Annif/wiki/Project-configuration>`_ |
|
158
|
|
|
for details. |
|
159
|
|
|
""" |
|
160
|
|
|
|
|
161
|
|
|
template = "{0: <25}{1: <45}{2: <10}{3: <7}" |
|
162
|
|
|
header = template.format( |
|
163
|
|
|
"Project ID", "Project Name", "Language", "Trained") |
|
164
|
|
|
click.echo(header) |
|
165
|
|
|
click.echo("-" * len(header)) |
|
166
|
|
|
for proj in annif.registry.get_projects( |
|
167
|
|
|
min_access=Access.private).values(): |
|
168
|
|
|
click.echo(template.format( |
|
169
|
|
|
proj.project_id, proj.name, proj.language, str(proj.is_trained))) |
|
170
|
|
|
|
|
171
|
|
|
|
|
172
|
|
|
@cli.command('show-project') |
|
173
|
|
|
@click.argument('project_id') |
|
174
|
|
|
@common_options |
|
175
|
|
|
def run_show_project(project_id): |
|
176
|
|
|
""" |
|
177
|
|
|
Show information about a project. |
|
178
|
|
|
""" |
|
179
|
|
|
|
|
180
|
|
|
proj = get_project(project_id) |
|
181
|
|
|
click.echo(f'Project ID: {proj.project_id}') |
|
182
|
|
|
click.echo(f'Project Name: {proj.name}') |
|
183
|
|
|
click.echo(f'Language: {proj.language}') |
|
184
|
|
|
click.echo(f'Vocabulary: {proj.vocab.vocab_id}') |
|
185
|
|
|
click.echo(f'Vocab language: {proj.vocab_lang}') |
|
186
|
|
|
click.echo(f'Access: {proj.access.name}') |
|
187
|
|
|
click.echo(f'Trained: {proj.is_trained}') |
|
188
|
|
|
click.echo(f'Modification time: {proj.modification_time}') |
|
189
|
|
|
|
|
190
|
|
|
|
|
191
|
|
|
@cli.command('clear') |
|
192
|
|
|
@click.argument('project_id') |
|
193
|
|
|
@common_options |
|
194
|
|
|
def run_clear_project(project_id): |
|
195
|
|
|
""" |
|
196
|
|
|
Initialize the project to its original, untrained state. |
|
197
|
|
|
""" |
|
198
|
|
|
proj = get_project(project_id) |
|
199
|
|
|
proj.remove_model_data() |
|
200
|
|
|
|
|
201
|
|
|
|
|
202
|
|
|
@cli.command('list-vocabs') |
|
203
|
|
|
@common_options |
|
204
|
|
|
@click_log.simple_verbosity_option(logger, default='ERROR') |
|
205
|
|
|
def run_list_vocabs(): |
|
206
|
|
|
""" |
|
207
|
|
|
List available vocabularies. |
|
208
|
|
|
""" |
|
209
|
|
|
|
|
210
|
|
|
template = "{0: <20}{1: <20}{2: >10} {3: <6}" |
|
211
|
|
|
header = template.format( |
|
212
|
|
|
"Vocabulary ID", "Languages", "Size", "Loaded") |
|
213
|
|
|
click.echo(header) |
|
214
|
|
|
click.echo("-" * len(header)) |
|
215
|
|
|
for vocab in annif.registry.get_vocabs( |
|
216
|
|
|
min_access=Access.private).values(): |
|
217
|
|
|
try: |
|
218
|
|
|
languages = ','.join(sorted(vocab.languages)) |
|
219
|
|
|
size = len(vocab) |
|
220
|
|
|
loaded = True |
|
221
|
|
|
except NotInitializedException: |
|
222
|
|
|
languages = '-' |
|
223
|
|
|
size = '-' |
|
224
|
|
|
loaded = False |
|
225
|
|
|
click.echo(template.format( |
|
226
|
|
|
vocab.vocab_id, languages, size, str(loaded))) |
|
227
|
|
|
|
|
228
|
|
|
|
|
229
|
|
|
@cli.command('load-vocab') |
|
230
|
|
|
@click.argument('vocab_id') |
|
231
|
|
|
@click.argument('subjectfile', type=click.Path(exists=True, dir_okay=False)) |
|
232
|
|
|
@click.option('--language', '-L', help='Language of subject file') |
|
233
|
|
|
@click.option('--force', '-f', default=False, is_flag=True, |
|
234
|
|
|
help='Replace existing vocabulary completely ' + |
|
235
|
|
|
'instead of updating it') |
|
236
|
|
|
@common_options |
|
237
|
|
|
def run_load_vocab(vocab_id, language, force, subjectfile): |
|
238
|
|
|
""" |
|
239
|
|
|
Load a vocabulary from a subject file. |
|
240
|
|
|
""" |
|
241
|
|
|
vocab = get_vocab(vocab_id) |
|
242
|
|
|
if annif.corpus.SubjectFileSKOS.is_rdf_file(subjectfile): |
|
243
|
|
|
# SKOS/RDF file supported by rdflib |
|
244
|
|
|
subjects = annif.corpus.SubjectFileSKOS(subjectfile) |
|
245
|
|
|
click.echo(f"Loading vocabulary from SKOS file {subjectfile}...") |
|
246
|
|
|
elif annif.corpus.SubjectFileCSV.is_csv_file(subjectfile): |
|
247
|
|
|
# CSV file |
|
248
|
|
|
subjects = annif.corpus.SubjectFileCSV(subjectfile) |
|
249
|
|
|
click.echo(f"Loading vocabulary from CSV file {subjectfile}...") |
|
250
|
|
|
else: |
|
251
|
|
|
# probably a TSV file - we need to know its language |
|
252
|
|
|
if not language: |
|
253
|
|
|
click.echo("Please use --language option to set the language of " + |
|
254
|
|
|
"a TSV vocabulary.", err=True) |
|
255
|
|
|
sys.exit(1) |
|
256
|
|
|
click.echo(f"Loading vocabulary from TSV file {subjectfile}...") |
|
257
|
|
|
subjects = annif.corpus.SubjectFileTSV(subjectfile, language) |
|
258
|
|
|
vocab.load_vocabulary(subjects, force=force) |
|
259
|
|
|
|
|
260
|
|
|
|
|
261
|
|
|
@cli.command('train') |
|
262
|
|
|
@click.argument('project_id') |
|
263
|
|
|
@click.argument('paths', type=click.Path(exists=True), nargs=-1) |
|
264
|
|
|
@click.option('--cached/--no-cached', '-c/-C', default=False, |
|
265
|
|
|
help='Reuse preprocessed training data from previous run') |
|
266
|
|
|
@click.option('--docs-limit', '-d', default=None, |
|
267
|
|
|
type=click.IntRange(0, None), |
|
268
|
|
|
help='Maximum number of documents to use') |
|
269
|
|
|
@click.option('--jobs', |
|
270
|
|
|
'-j', |
|
271
|
|
|
default=0, |
|
272
|
|
|
help='Number of parallel jobs (0 means choose automatically)') |
|
273
|
|
|
@backend_param_option |
|
274
|
|
|
@common_options |
|
275
|
|
|
def run_train(project_id, paths, cached, docs_limit, jobs, backend_param): |
|
276
|
|
|
""" |
|
277
|
|
|
Train a project on a collection of documents. |
|
278
|
|
|
\f |
|
279
|
|
|
This will train the project using the documents from ``PATHS`` (directories |
|
280
|
|
|
or possibly gzipped TSV files) in a single batch operation. If ``--cached`` |
|
281
|
|
|
is set, preprocessed training data from the previous run is reused instead |
|
282
|
|
|
of documents input; see `Reusing preprocessed training data |
|
283
|
|
|
<https://github.com/NatLibFi/Annif/wiki/ |
|
284
|
|
|
Reusing-preprocessed-training-data>`_. |
|
285
|
|
|
""" |
|
286
|
|
|
proj = get_project(project_id) |
|
287
|
|
|
backend_params = parse_backend_params(backend_param, proj) |
|
288
|
|
|
if cached: |
|
289
|
|
|
if len(paths) > 0: |
|
290
|
|
|
raise click.UsageError( |
|
291
|
|
|
"Corpus paths cannot be given when using --cached option.") |
|
292
|
|
|
documents = 'cached' |
|
293
|
|
|
else: |
|
294
|
|
|
documents = open_documents(paths, proj.subjects, |
|
295
|
|
|
proj.vocab_lang, docs_limit) |
|
296
|
|
|
proj.train(documents, backend_params, jobs) |
|
297
|
|
|
|
|
298
|
|
|
|
|
299
|
|
|
@cli.command('learn') |
|
300
|
|
|
@click.argument('project_id') |
|
301
|
|
|
@click.argument('paths', type=click.Path(exists=True), nargs=-1) |
|
302
|
|
|
@click.option('--docs-limit', '-d', default=None, |
|
303
|
|
|
type=click.IntRange(0, None), |
|
304
|
|
|
help='Maximum number of documents to use') |
|
305
|
|
|
@backend_param_option |
|
306
|
|
|
@common_options |
|
307
|
|
|
def run_learn(project_id, paths, docs_limit, backend_param): |
|
308
|
|
|
""" |
|
309
|
|
|
Further train an existing project on a collection of documents. |
|
310
|
|
|
\f |
|
311
|
|
|
Similar to the ``train`` command. This will continue training an already |
|
312
|
|
|
trained project using the documents given by ``PATHS`` in a single batch |
|
313
|
|
|
operation. Not supported by all backends. |
|
314
|
|
|
""" |
|
315
|
|
|
proj = get_project(project_id) |
|
316
|
|
|
backend_params = parse_backend_params(backend_param, proj) |
|
317
|
|
|
documents = open_documents(paths, proj.subjects, |
|
318
|
|
|
proj.vocab_lang, docs_limit) |
|
319
|
|
|
proj.learn(documents, backend_params) |
|
320
|
|
|
|
|
321
|
|
|
|
|
322
|
|
|
@cli.command('suggest') |
|
323
|
|
|
@click.argument('project_id') |
|
324
|
|
|
@click.option('--limit', '-l', default=10, help='Maximum number of subjects') |
|
325
|
|
|
@click.option('--threshold', '-t', default=0.0, help='Minimum score threshold') |
|
326
|
|
|
@click.option('--language', '-L', help='Language of subject labels') |
|
327
|
|
|
@backend_param_option |
|
328
|
|
|
@common_options |
|
329
|
|
|
def run_suggest(project_id, limit, threshold, language, backend_param): |
|
330
|
|
|
""" |
|
331
|
|
|
Suggest subjects for a single document from standard input. |
|
332
|
|
|
\f |
|
333
|
|
|
This will read a text document from standard input and suggest subjects for |
|
334
|
|
|
it. |
|
335
|
|
|
""" |
|
336
|
|
|
project = get_project(project_id) |
|
337
|
|
|
text = sys.stdin.read() |
|
338
|
|
|
lang = language or project.vocab_lang |
|
339
|
|
|
if lang not in project.vocab.languages: |
|
340
|
|
|
raise click.BadParameter( |
|
341
|
|
|
f'language "{lang}" not supported by vocabulary') |
|
342
|
|
|
backend_params = parse_backend_params(backend_param, project) |
|
343
|
|
|
hit_filter = SuggestionFilter(project.subjects, limit, threshold) |
|
344
|
|
|
hits = hit_filter(project.suggest(text, backend_params)) |
|
345
|
|
|
for hit in hits.as_list(): |
|
346
|
|
|
subj = project.subjects[hit.subject_id] |
|
347
|
|
|
click.echo( |
|
348
|
|
|
"<{}>\t{}\t{}".format( |
|
349
|
|
|
subj.uri, |
|
350
|
|
|
'\t'.join(filter(None, |
|
351
|
|
|
(subj.labels[lang], |
|
352
|
|
|
subj.notation))), |
|
353
|
|
|
hit.score)) |
|
354
|
|
|
|
|
355
|
|
|
|
|
356
|
|
|
@cli.command('index') |
|
357
|
|
|
@click.argument('project_id') |
|
358
|
|
|
@click.argument('directory', type=click.Path(exists=True, file_okay=False)) |
|
359
|
|
|
@click.option( |
|
360
|
|
|
'--suffix', |
|
361
|
|
|
'-s', |
|
362
|
|
|
default='.annif', |
|
363
|
|
|
help='File name suffix for result files') |
|
364
|
|
|
@click.option('--force/--no-force', '-f/-F', default=False, |
|
365
|
|
|
help='Force overwriting of existing result files') |
|
366
|
|
|
@click.option('--limit', '-l', default=10, help='Maximum number of subjects') |
|
367
|
|
|
@click.option('--threshold', '-t', default=0.0, help='Minimum score threshold') |
|
368
|
|
|
@click.option('--language', '-L', help='Language of subject labels') |
|
369
|
|
|
@backend_param_option |
|
370
|
|
|
@common_options |
|
371
|
|
|
def run_index(project_id, directory, suffix, force, |
|
372
|
|
|
limit, threshold, language, backend_param): |
|
373
|
|
|
""" |
|
374
|
|
|
Index a directory with documents, suggesting subjects for each document. |
|
375
|
|
|
Write the results in TSV files with the given suffix (``.annif`` by |
|
376
|
|
|
default). |
|
377
|
|
|
""" |
|
378
|
|
|
project = get_project(project_id) |
|
379
|
|
|
lang = language or project.vocab_lang |
|
380
|
|
|
if lang not in project.vocab.languages: |
|
381
|
|
|
raise click.BadParameter( |
|
382
|
|
|
f'language "{lang}" not supported by vocabulary') |
|
383
|
|
|
backend_params = parse_backend_params(backend_param, project) |
|
384
|
|
|
hit_filter = SuggestionFilter(project.subjects, limit, threshold) |
|
385
|
|
|
|
|
386
|
|
|
for docfilename, dummy_subjectfn in annif.corpus.DocumentDirectory( |
|
387
|
|
|
directory, project.subjects, project.vocab_lang, |
|
388
|
|
|
require_subjects=False): |
|
389
|
|
|
with open(docfilename, encoding='utf-8-sig') as docfile: |
|
390
|
|
|
text = docfile.read() |
|
391
|
|
|
subjectfilename = re.sub(r'\.txt$', suffix, docfilename) |
|
392
|
|
|
if os.path.exists(subjectfilename) and not force: |
|
393
|
|
|
click.echo( |
|
394
|
|
|
"Not overwriting {} (use --force to override)".format( |
|
395
|
|
|
subjectfilename)) |
|
396
|
|
|
continue |
|
397
|
|
|
with open(subjectfilename, 'w', encoding='utf-8') as subjfile: |
|
398
|
|
|
results = project.suggest(text, backend_params) |
|
399
|
|
|
for hit in hit_filter(results).as_list(): |
|
400
|
|
|
subj = project.subjects[hit.subject_id] |
|
401
|
|
|
line = "<{}>\t{}\t{}".format( |
|
402
|
|
|
subj.uri, |
|
403
|
|
|
'\t'.join(filter(None, (subj.labels[lang], |
|
404
|
|
|
subj.notation))), |
|
405
|
|
|
hit.score) |
|
406
|
|
|
click.echo(line, file=subjfile) |
|
407
|
|
|
|
|
408
|
|
|
|
|
409
|
|
|
@cli.command('eval') |
|
410
|
|
|
@click.argument('project_id') |
|
411
|
|
|
@click.argument('paths', type=click.Path(exists=True), nargs=-1) |
|
412
|
|
|
@click.option('--limit', '-l', default=10, help='Maximum number of subjects') |
|
413
|
|
|
@click.option('--threshold', '-t', default=0.0, help='Minimum score threshold') |
|
414
|
|
|
@click.option('--docs-limit', '-d', default=None, |
|
415
|
|
|
type=click.IntRange(0, None), |
|
416
|
|
|
help='Maximum number of documents to use') |
|
417
|
|
|
@click.option('--metric', '-m', default=[], multiple=True, |
|
418
|
|
|
help='Metric to calculate (default: all)') |
|
419
|
|
|
@click.option( |
|
420
|
|
|
'--metrics-file', |
|
421
|
|
|
'-M', |
|
422
|
|
|
type=click.File( |
|
423
|
|
|
'w', |
|
424
|
|
|
encoding='utf-8', |
|
425
|
|
|
errors='ignore', |
|
426
|
|
|
lazy=True), |
|
427
|
|
|
help="""Specify file in order to write evaluation metrics in JSON format. |
|
428
|
|
|
File directory must exist, existing file will be overwritten.""") |
|
429
|
|
|
@click.option( |
|
430
|
|
|
'--results-file', |
|
431
|
|
|
'-r', |
|
432
|
|
|
type=click.File( |
|
433
|
|
|
'w', |
|
434
|
|
|
encoding='utf-8', |
|
435
|
|
|
errors='ignore', |
|
436
|
|
|
lazy=True), |
|
437
|
|
|
help="""Specify file in order to write non-aggregated results per subject. |
|
438
|
|
|
File directory must exist, existing file will be overwritten.""") |
|
439
|
|
|
@click.option('--jobs', |
|
440
|
|
|
'-j', |
|
441
|
|
|
default=1, |
|
442
|
|
|
help='Number of parallel jobs (0 means all CPUs)') |
|
443
|
|
|
@backend_param_option |
|
444
|
|
|
@common_options |
|
445
|
|
|
def run_eval( |
|
446
|
|
|
project_id, |
|
447
|
|
|
paths, |
|
448
|
|
|
limit, |
|
449
|
|
|
threshold, |
|
450
|
|
|
docs_limit, |
|
451
|
|
|
metric, |
|
452
|
|
|
metrics_file, |
|
453
|
|
|
results_file, |
|
454
|
|
|
jobs, |
|
455
|
|
|
backend_param): |
|
456
|
|
|
""" |
|
457
|
|
|
Suggest subjects for documents and evaluate the results by comparing |
|
458
|
|
|
against a gold standard. |
|
459
|
|
|
\f |
|
460
|
|
|
With this command the documents from ``PATHS`` (directories or possibly |
|
461
|
|
|
gzipped TSV files) will be assigned subject suggestions and then |
|
462
|
|
|
statistical measures are calculated that quantify how well the suggested |
|
463
|
|
|
subjects match the gold-standard subjects in the documents. |
|
464
|
|
|
|
|
465
|
|
|
Normally the output is the list of the metrics calculated across documents. |
|
466
|
|
|
If ``--results-file <FILENAME>`` option is given, the metrics are |
|
467
|
|
|
calculated separately for each subject, and written to the given file. |
|
468
|
|
|
""" |
|
469
|
|
|
|
|
470
|
|
|
project = get_project(project_id) |
|
471
|
|
|
backend_params = parse_backend_params(backend_param, project) |
|
472
|
|
|
|
|
473
|
|
|
import annif.eval |
|
474
|
|
|
eval_batch = annif.eval.EvaluationBatch(project.subjects) |
|
475
|
|
|
|
|
476
|
|
|
if results_file: |
|
477
|
|
|
try: |
|
478
|
|
|
print('', end='', file=results_file) |
|
479
|
|
|
click.echo('Writing per subject evaluation results to {!s}'.format( |
|
480
|
|
|
results_file.name)) |
|
481
|
|
|
except Exception as e: |
|
482
|
|
|
raise NotSupportedException( |
|
483
|
|
|
"cannot open results-file for writing: " + str(e)) |
|
484
|
|
|
docs = open_documents(paths, project.subjects, |
|
485
|
|
|
project.vocab_lang, docs_limit) |
|
486
|
|
|
|
|
487
|
|
|
jobs, pool_class = annif.parallel.get_pool(jobs) |
|
488
|
|
|
|
|
489
|
|
|
project.initialize(parallel=True) |
|
490
|
|
|
psmap = annif.parallel.ProjectSuggestMap( |
|
491
|
|
|
project.registry, [project_id], backend_params, limit, threshold) |
|
492
|
|
|
|
|
493
|
|
|
with pool_class(jobs) as pool: |
|
494
|
|
|
for hits, subject_set in pool.imap_unordered( |
|
495
|
|
|
psmap.suggest, docs.documents): |
|
496
|
|
|
eval_batch.evaluate(hits[project_id], |
|
497
|
|
|
subject_set) |
|
498
|
|
|
|
|
499
|
|
|
template = "{0:<30}\t{1}" |
|
500
|
|
|
metrics = eval_batch.results(metrics=metric, |
|
501
|
|
|
results_file=results_file, |
|
502
|
|
|
language=project.vocab_lang) |
|
503
|
|
|
for metric, score in metrics.items(): |
|
504
|
|
|
click.echo(template.format(metric + ":", score)) |
|
505
|
|
|
if metrics_file: |
|
506
|
|
|
json.dump( |
|
507
|
|
|
{metric_code(mname): val for mname, val in metrics.items()}, |
|
508
|
|
|
metrics_file, indent=2) |
|
509
|
|
|
|
|
510
|
|
|
|
|
511
|
|
|
@cli.command('optimize') |
|
512
|
|
|
@click.argument('project_id') |
|
513
|
|
|
@click.argument('paths', type=click.Path(exists=True), nargs=-1) |
|
514
|
|
|
@click.option('--docs-limit', '-d', default=None, |
|
515
|
|
|
type=click.IntRange(0, None), |
|
516
|
|
|
help='Maximum number of documents to use') |
|
517
|
|
|
@backend_param_option |
|
518
|
|
|
@common_options |
|
519
|
|
|
def run_optimize(project_id, paths, docs_limit, backend_param): |
|
520
|
|
|
""" |
|
521
|
|
|
Suggest subjects for documents, testing multiple limits and thresholds. |
|
522
|
|
|
\f |
|
523
|
|
|
This command will use different limit (maximum number of subjects) and |
|
524
|
|
|
score threshold values when assigning subjects to each document given by |
|
525
|
|
|
``PATHS`` and compare the results against the gold standard subjects in the |
|
526
|
|
|
documents. The output is a list of parameter combinations and their scores. |
|
527
|
|
|
From the output, you can determine the optimum limit and threshold |
|
528
|
|
|
parameters depending on which measure you want to target. |
|
529
|
|
|
""" |
|
530
|
|
|
project = get_project(project_id) |
|
531
|
|
|
backend_params = parse_backend_params(backend_param, project) |
|
532
|
|
|
|
|
533
|
|
|
filter_batches = generate_filter_batches(project.subjects) |
|
534
|
|
|
|
|
535
|
|
|
ndocs = 0 |
|
536
|
|
|
docs = open_documents(paths, project.subjects, |
|
537
|
|
|
project.vocab_lang, docs_limit) |
|
538
|
|
|
for doc in docs.documents: |
|
539
|
|
|
raw_hits = project.suggest(doc.text, backend_params) |
|
540
|
|
|
hits = raw_hits.filter(project.subjects, limit=BATCH_MAX_LIMIT) |
|
541
|
|
|
assert isinstance(hits, ListSuggestionResult), \ |
|
542
|
|
|
"Optimize should only be done with ListSuggestionResult " + \ |
|
543
|
|
|
"as it would be very slow with VectorSuggestionResult." |
|
544
|
|
|
for hit_filter, batch in filter_batches.values(): |
|
545
|
|
|
batch.evaluate(hit_filter(hits), doc.subject_set) |
|
546
|
|
|
ndocs += 1 |
|
547
|
|
|
|
|
548
|
|
|
click.echo("\t".join(('Limit', 'Thresh.', 'Prec.', 'Rec.', 'F1'))) |
|
549
|
|
|
|
|
550
|
|
|
best_scores = collections.defaultdict(float) |
|
551
|
|
|
best_params = {} |
|
552
|
|
|
|
|
553
|
|
|
template = "{:d}\t{:.02f}\t{:.04f}\t{:.04f}\t{:.04f}" |
|
554
|
|
|
# Store the batches in a list that gets consumed along the way |
|
555
|
|
|
# This way GC will have a chance to reclaim the memory |
|
556
|
|
|
filter_batches = list(filter_batches.items()) |
|
557
|
|
|
while filter_batches: |
|
558
|
|
|
params, filter_batch = filter_batches.pop(0) |
|
559
|
|
|
metrics = ['Precision (doc avg)', |
|
560
|
|
|
'Recall (doc avg)', |
|
561
|
|
|
'F1 score (doc avg)'] |
|
562
|
|
|
results = filter_batch[1].results(metrics=metrics) |
|
563
|
|
|
for metric, score in results.items(): |
|
564
|
|
|
if score >= best_scores[metric]: |
|
565
|
|
|
best_scores[metric] = score |
|
566
|
|
|
best_params[metric] = params |
|
567
|
|
|
click.echo( |
|
568
|
|
|
template.format( |
|
569
|
|
|
params[0], |
|
570
|
|
|
params[1], |
|
571
|
|
|
results['Precision (doc avg)'], |
|
572
|
|
|
results['Recall (doc avg)'], |
|
573
|
|
|
results['F1 score (doc avg)'])) |
|
574
|
|
|
|
|
575
|
|
|
click.echo() |
|
576
|
|
|
template2 = "Best {:>19}: {:.04f}\tLimit: {:d}\tThreshold: {:.02f}" |
|
577
|
|
|
for metric in metrics: |
|
578
|
|
|
click.echo( |
|
579
|
|
|
template2.format( |
|
580
|
|
|
metric, |
|
581
|
|
|
best_scores[metric], |
|
582
|
|
|
best_params[metric][0], |
|
583
|
|
|
best_params[metric][1])) |
|
584
|
|
|
click.echo("Documents evaluated:\t{}".format(ndocs)) |
|
585
|
|
|
|
|
586
|
|
|
|
|
587
|
|
|
@cli.command('hyperopt') |
|
588
|
|
|
@click.argument('project_id') |
|
589
|
|
|
@click.argument('paths', type=click.Path(exists=True), nargs=-1) |
|
590
|
|
|
@click.option('--docs-limit', '-d', default=None, |
|
591
|
|
|
type=click.IntRange(0, None), |
|
592
|
|
|
help='Maximum number of documents to use') |
|
593
|
|
|
@click.option('--trials', '-T', default=10, help='Number of trials') |
|
594
|
|
|
@click.option('--jobs', |
|
595
|
|
|
'-j', |
|
596
|
|
|
default=1, |
|
597
|
|
|
help='Number of parallel runs (0 means all CPUs)') |
|
598
|
|
|
@click.option('--metric', '-m', default='NDCG', |
|
599
|
|
|
help='Metric to optimize (default: NDCG)') |
|
600
|
|
|
@click.option( |
|
601
|
|
|
'--results-file', |
|
602
|
|
|
'-r', |
|
603
|
|
|
type=click.File( |
|
604
|
|
|
'w', |
|
605
|
|
|
encoding='utf-8', |
|
606
|
|
|
errors='ignore', |
|
607
|
|
|
lazy=True), |
|
608
|
|
|
help="""Specify file path to write trial results as CSV. |
|
609
|
|
|
File directory must exist, existing file will be overwritten.""") |
|
610
|
|
|
@common_options |
|
611
|
|
|
def run_hyperopt(project_id, paths, docs_limit, trials, jobs, metric, |
|
612
|
|
|
results_file): |
|
613
|
|
|
""" |
|
614
|
|
|
Optimize the hyperparameters of a project using validation documents from |
|
615
|
|
|
``PATHS``. Not supported by all backends. Output is a list of trial results |
|
616
|
|
|
and a report of the best performing parameters. |
|
617
|
|
|
""" |
|
618
|
|
|
proj = get_project(project_id) |
|
619
|
|
|
documents = open_documents(paths, proj.subjects, |
|
620
|
|
|
proj.vocab_lang, docs_limit) |
|
621
|
|
|
click.echo(f"Looking for optimal hyperparameters using {trials} trials") |
|
622
|
|
|
rec = proj.hyperopt(documents, trials, jobs, metric, results_file) |
|
623
|
|
|
click.echo(f"Got best {metric} score {rec.score:.4f} with:") |
|
624
|
|
|
click.echo("---") |
|
625
|
|
|
for line in rec.lines: |
|
626
|
|
|
click.echo(line) |
|
627
|
|
|
click.echo("---") |
|
628
|
|
|
|
|
629
|
|
|
|
|
630
|
|
|
if __name__ == '__main__': |
|
631
|
|
|
cli() |
|
632
|
|
|
|
NatLibFi /
Annif
