|
a |
|
b/tests/evaluation/test_evaluate.py |
|
|
1 |
from deidentify.base import Annotation, Document |
|
|
2 |
from deidentify.evaluation.evaluator import ENTITY_TAG, Evaluator |
|
|
3 |
|
|
|
4 |
|
|
|
5 |
def test_entity_level(): |
|
|
6 |
gold = [ |
|
|
7 |
Document(name='doc_a', text='', annotations=[Annotation('', 3, 6, 'MISC')]), |
|
|
8 |
Document(name='doc_b', text='', annotations=[Annotation('', 0, 2, 'PER')]) |
|
|
9 |
] |
|
|
10 |
|
|
|
11 |
predicted = [ |
|
|
12 |
Document(name='doc_a', text='', annotations=[Annotation('', 2, 6, 'MISC')]), |
|
|
13 |
Document(name='doc_b', text='', annotations=[Annotation('', 0, 2, 'PER')]) |
|
|
14 |
] |
|
|
15 |
|
|
|
16 |
evaluator = Evaluator(gold, predicted) |
|
|
17 |
scores = evaluator.entity_level() |
|
|
18 |
assert scores.micro_avg_f_score() == 0.5 |
|
|
19 |
assert scores.macro_avg_f_score() == 0.5 |
|
|
20 |
assert scores.f_score('PER') == 1 |
|
|
21 |
assert scores.f_score('MISC') == 0 |
|
|
22 |
|
|
|
23 |
|
|
|
24 |
def test_token_annotations(): |
|
|
25 |
evaluator = Evaluator(gold=(), predicted=()) |
|
|
26 |
doc = Document(name='doc_a', text='A B C D.', annotations=[ |
|
|
27 |
Annotation('B C', 2, 5, 'PER'), |
|
|
28 |
Annotation('D.', 6, 8, 'ORG') |
|
|
29 |
]) |
|
|
30 |
|
|
|
31 |
assert evaluator.token_annotations(doc) == ['O', 'PER', 'PER', 'ORG'] |
|
|
32 |
assert evaluator.token_annotations(doc, tag_blind=True) == ['O', 'ENT', 'ENT', 'ENT'] |
|
|
33 |
|
|
|
34 |
|
|
|
35 |
def test_token_level(): |
|
|
36 |
text = 'A B C D.' |
|
|
37 |
|
|
|
38 |
gold_a = [Annotation('B C', 2, 5, 'PER')] |
|
|
39 |
gold_b = [Annotation('A', 0, 1, 'ORG'), Annotation('B', 2, 3, 'PER')] |
|
|
40 |
|
|
|
41 |
pred_a = [Annotation('B', 2, 3, 'PER'), Annotation('C', 4, 5, 'PER')] |
|
|
42 |
pred_b = [Annotation('A', 0, 1, 'ORG'), Annotation('B', 2, 3, 'ORG')] |
|
|
43 |
|
|
|
44 |
gold = [ |
|
|
45 |
Document(name='doc_a', text=text, annotations=gold_a), |
|
|
46 |
Document(name='doc_b', text=text, annotations=gold_b) |
|
|
47 |
] |
|
|
48 |
|
|
|
49 |
predicted = [ |
|
|
50 |
Document(name='doc_a', text=text, annotations=pred_a), |
|
|
51 |
Document(name='doc_b', text=text, annotations=pred_b) |
|
|
52 |
] |
|
|
53 |
|
|
|
54 |
evaluator = Evaluator(gold, predicted) |
|
|
55 |
scores = evaluator.token_level() |
|
|
56 |
assert scores.precision('PER') == 1 |
|
|
57 |
assert scores.recall('PER') == 0.6667 |
|
|
58 |
assert scores.f_score('PER') == 0.8 |
|
|
59 |
|
|
|
60 |
assert scores.precision('ORG') == 0.5 |
|
|
61 |
assert scores.recall('ORG') == 1 |
|
|
62 |
assert scores.f_score('ORG') == 0.6667 |
|
|
63 |
|
|
|
64 |
|
|
|
65 |
def test_token_level_blind(): |
|
|
66 |
gold_a = [Annotation('B C', 2, 5, 'PER')] |
|
|
67 |
gold_b = [Annotation('A', 0, 1, 'ORG')] |
|
|
68 |
|
|
|
69 |
pred_a = [Annotation('B', 2, 3, 'PER'), Annotation('C', 4, 5, 'PER')] |
|
|
70 |
pred_b = [] |
|
|
71 |
|
|
|
72 |
gold = [ |
|
|
73 |
Document(name='doc_a', text='A B C D.', annotations=gold_a), |
|
|
74 |
Document(name='doc_b', text='A B C D.', annotations=gold_b) |
|
|
75 |
] |
|
|
76 |
|
|
|
77 |
predicted = [ |
|
|
78 |
Document(name='doc_a', text='A B C D.', annotations=pred_a), |
|
|
79 |
Document(name='doc_b', text='A B C D.', annotations=pred_b) |
|
|
80 |
] |
|
|
81 |
|
|
|
82 |
evaluator = Evaluator(gold, predicted) |
|
|
83 |
scores = evaluator.token_level_blind() |
|
|
84 |
assert scores.precision(ENTITY_TAG) == 1 |
|
|
85 |
assert scores.recall(ENTITY_TAG) == 0.6667 |
|
|
86 |
assert scores.f_score(ENTITY_TAG) == 0.8 |