|
1
|
|
|
"""Parallel processing functionality for Annif""" |
|
2
|
|
|
|
|
3
|
|
|
|
|
4
|
|
|
import multiprocessing |
|
5
|
|
|
import multiprocessing.dummy |
|
6
|
|
|
from collections import defaultdict |
|
7
|
|
|
|
|
8
|
|
|
# Start method for processes created by the multiprocessing module. |
|
9
|
|
|
# A value of None means using the platform-specific default. |
|
10
|
|
|
# Intended to be overridden in unit tests. |
|
11
|
|
|
MP_START_METHOD = None |
|
12
|
|
|
|
|
13
|
|
|
|
|
14
|
|
|
class BaseWorker: |
|
15
|
|
|
"""Base class for workers that implement tasks executed via |
|
16
|
|
|
multiprocessing. The init method can be used to store data objects that |
|
17
|
|
|
are necessary for the operation. They will be stored in a class |
|
18
|
|
|
attribute that is accessible to the static worker method. The storage |
|
19
|
|
|
solution is inspired by this blog post: |
|
20
|
|
|
https://thelaziestprogrammer.com/python/multiprocessing-pool-a-global-solution # noqa |
|
21
|
|
|
""" |
|
22
|
|
|
|
|
23
|
|
|
args = None |
|
24
|
|
|
|
|
25
|
|
|
@classmethod |
|
26
|
|
|
def init(cls, args): |
|
27
|
|
|
cls.args = args # pragma: no cover |
|
28
|
|
|
|
|
29
|
|
|
|
|
30
|
|
|
class ProjectSuggestMap: |
|
31
|
|
|
"""A utility class that can be used to wrap one or more projects and |
|
32
|
|
|
provide a mapping method that converts Document objects to suggestions. |
|
33
|
|
|
Intended to be used with the multiprocessing module.""" |
|
34
|
|
|
|
|
35
|
|
|
def __init__(self, registry, project_ids, backend_params, limit, threshold): |
|
36
|
|
|
self.registry = registry |
|
37
|
|
|
self.project_ids = project_ids |
|
38
|
|
|
self.backend_params = backend_params |
|
39
|
|
|
self.limit = limit |
|
40
|
|
|
self.threshold = threshold |
|
41
|
|
|
|
|
42
|
|
|
def suggest_batch(self, batch): |
|
43
|
|
|
filtered_hit_sets = defaultdict(list) |
|
44
|
|
|
texts, subject_sets = zip(*[(doc.text, doc.subject_set) for doc in batch]) |
|
45
|
|
|
|
|
46
|
|
|
for project_id in self.project_ids: |
|
47
|
|
|
project = self.registry.get_project(project_id) |
|
48
|
|
|
hit_sets = project.suggest(texts, self.backend_params) |
|
49
|
|
|
for hits in hit_sets: |
|
50
|
|
|
filtered_hit_sets[project_id].append( |
|
51
|
|
|
hits.filter(project.subjects, self.limit, self.threshold) |
|
52
|
|
|
) |
|
53
|
|
|
return (filtered_hit_sets, subject_sets) |
|
54
|
|
|
|
|
55
|
|
|
|
|
56
|
|
|
def get_pool(n_jobs): |
|
57
|
|
|
"""return a suitable multiprocessing pool class, and the correct jobs |
|
58
|
|
|
argument for its constructor, for the given amount of parallel jobs""" |
|
59
|
|
|
|
|
60
|
|
|
ctx = multiprocessing.get_context(MP_START_METHOD) |
|
61
|
|
|
|
|
62
|
|
|
if n_jobs < 1: |
|
63
|
|
|
n_jobs = None |
|
64
|
|
|
pool_class = ctx.Pool |
|
65
|
|
|
elif n_jobs == 1: |
|
66
|
|
|
# use the dummy wrapper around threading to avoid subprocess overhead |
|
67
|
|
|
pool_class = multiprocessing.dummy.Pool |
|
68
|
|
|
else: |
|
69
|
|
|
pool_class = ctx.Pool |
|
70
|
|
|
|
|
71
|
|
|
return n_jobs, pool_class |
|
72
|
|
|
|
NatLibFi /
Annif
