1
|
|
|
"""Parallel processing functionality for Annif""" |
2
|
|
|
|
3
|
|
|
from __future__ import annotations |
4
|
|
|
|
5
|
|
|
import multiprocessing |
6
|
|
|
import multiprocessing.dummy |
7
|
|
|
from typing import TYPE_CHECKING, Any |
8
|
|
|
|
9
|
|
|
if TYPE_CHECKING: |
10
|
|
|
from collections import defaultdict |
11
|
|
|
from collections.abc import Iterator |
12
|
|
|
from typing import Callable |
13
|
|
|
|
14
|
|
|
from annif.corpus import Document, SubjectSet |
15
|
|
|
from annif.registry import AnnifRegistry |
16
|
|
|
from annif.suggestion import SuggestionBatch, SuggestionResult |
17
|
|
|
|
18
|
|
|
|
19
|
|
|
# Start method for processes created by the multiprocessing module. |
20
|
|
|
# A value of None means using the platform-specific default. |
21
|
|
|
# Intended to be overridden in unit tests. |
22
|
|
|
MP_START_METHOD = None |
23
|
|
|
|
24
|
|
|
|
25
|
|
|
class BaseWorker: |
26
|
|
|
"""Base class for workers that implement tasks executed via |
27
|
|
|
multiprocessing. The init method can be used to store data objects that |
28
|
|
|
are necessary for the operation. They will be stored in a class |
29
|
|
|
attribute that is accessible to the static worker method. The storage |
30
|
|
|
solution is inspired by this blog post: |
31
|
|
|
https://thelaziestprogrammer.com/python/multiprocessing-pool-a-global-solution # noqa |
32
|
|
|
""" |
33
|
|
|
|
34
|
|
|
args = None |
35
|
|
|
|
36
|
|
|
@classmethod |
37
|
|
|
def init(cls, args) -> None: |
38
|
|
|
cls.args = args # pragma: no cover |
39
|
|
|
|
40
|
|
|
|
41
|
|
|
class ProjectSuggestMap: |
42
|
|
|
"""A utility class that can be used to wrap one or more projects and |
43
|
|
|
provide a mapping method that converts Document objects to suggestions. |
44
|
|
|
Intended to be used with the multiprocessing module.""" |
45
|
|
|
|
46
|
|
|
def __init__( |
47
|
|
|
self, |
48
|
|
|
registry: AnnifRegistry, |
49
|
|
|
project_ids: list[str], |
50
|
|
|
backend_params: defaultdict[str, Any] | None, |
51
|
|
|
limit: int | None, |
52
|
|
|
threshold: float, |
53
|
|
|
) -> None: |
54
|
|
|
self.registry = registry |
55
|
|
|
self.project_ids = project_ids |
56
|
|
|
self.backend_params = backend_params |
57
|
|
|
self.limit = limit |
58
|
|
|
self.threshold = threshold |
59
|
|
|
|
60
|
|
|
def suggest(self, doc: Document) -> tuple[dict[str, SuggestionResult], SubjectSet]: |
61
|
|
|
filtered_hits = {} |
62
|
|
|
for project_id in self.project_ids: |
63
|
|
|
project = self.registry.get_project(project_id) |
64
|
|
|
batch = project.suggest([doc], self.backend_params) |
65
|
|
|
filtered_hits[project_id] = batch.filter(self.limit, self.threshold)[0] |
66
|
|
|
return (filtered_hits, doc.subject_set) |
67
|
|
|
|
68
|
|
|
def suggest_batch( |
69
|
|
|
self, batch |
70
|
|
|
) -> tuple[dict[str, SuggestionBatch], Iterator[SubjectSet]]: |
71
|
|
|
filtered_hit_sets = {} |
72
|
|
|
subject_sets = [doc.subject_set for doc in batch] |
73
|
|
|
|
74
|
|
|
for project_id in self.project_ids: |
75
|
|
|
project = self.registry.get_project(project_id) |
76
|
|
|
suggestion_batch = project.suggest(batch, self.backend_params) |
77
|
|
|
filtered_hit_sets[project_id] = suggestion_batch.filter( |
78
|
|
|
self.limit, self.threshold |
79
|
|
|
) |
80
|
|
|
return (filtered_hit_sets, subject_sets) |
81
|
|
|
|
82
|
|
|
|
83
|
|
|
def get_pool(n_jobs: int) -> tuple[int | None, Callable]: |
84
|
|
|
"""return a suitable constructor for multiprocessing pool class, and the correct |
85
|
|
|
jobs argument for it, for the given amount of parallel jobs""" |
86
|
|
|
|
87
|
|
|
ctx = multiprocessing.get_context(MP_START_METHOD) |
88
|
|
|
|
89
|
|
|
if n_jobs < 1: |
90
|
|
|
n_jobs = None |
91
|
|
|
pool_constructor: Callable = ctx.Pool |
|
|
|
|
92
|
|
|
elif n_jobs == 1: |
93
|
|
|
# use the dummy wrapper around threading to avoid subprocess overhead |
94
|
|
|
pool_constructor = multiprocessing.dummy.Pool |
95
|
|
|
else: |
96
|
|
|
pool_constructor = ctx.Pool |
97
|
|
|
|
98
|
|
|
return n_jobs, pool_constructor |
99
|
|
|
|