-
Notifications
You must be signed in to change notification settings - Fork 5
/
utils.py
170 lines (147 loc) · 6.67 KB
/
utils.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
import numpy as np
import pandas as pd
from typing import Any, Union, Dict, Iterable, List, Optional, Tuple
from sklearn.metrics import confusion_matrix,classification_report
from metrics import NSDSpanBasedF1Measure
def confidence(features: np.ndarray,
means: np.ndarray,
distance_type: str,
cov: np.ndarray = None) -> np.ndarray:
"""
Calculate mahalanobis or euclidean based confidence score for each class.
Params:
- features: shape (num_samples, num_features)
- means: shape (num_classes, num_features)
- cov: shape (num_features, num_features) or None (if use euclidean distance)
Returns:
- confidence: shape (num_samples, num_classes)
"""
assert distance_type in ("euclidean", "mahalanobis")
num_samples = features.shape[0]
num_features = features.shape[1]
num_classes = means.shape[0]
if distance_type == "euclidean":
cov = np.identity(num_features)
features = features.reshape(num_samples, 1, num_features).repeat(num_classes,axis=1) # (num_samples, num_classes, num_features)
means = means.reshape(1, num_classes, num_features).repeat(num_samples,axis=0) # (num_samples, num_classes, num_features)
vectors = features - means # (num_samples, num_classes, num_features)
cov_inv = np.linalg.inv(cov)
bef_sqrt = np.matmul(np.matmul(vectors.reshape(num_samples, num_classes, 1, num_features), cov_inv),
vectors.reshape(num_samples, num_classes, num_features, 1)).squeeze()
result = np.sqrt(bef_sqrt)
result[np.isnan(result)] = 1e12 # solve nan
return result
def parse_line(line:List): # [for Spanf1]
"""
Given the predicted sequence (contain "ns"), return the parsed BIO sequence (contain "B-ns" and "I-ns").
And due to the override, IND tags may be discordant, so we need to adjust the IND tags after been override.
e.g.
Input : ns ns ns O B-playlist_owner B-playlist ns I-playlist O
Return : B-ns I-ns I-ns O B-playlist_owner B-playlist B-ns B-playlist O
Params:
- line(list),the prediced sequence (contain "ns").
Returns:
- adjust_line(list), the parsed BIO sequence(contain "B-ns" and "I-ns")
"""
adjust_line = []
for i,label in enumerate(line):
if label in ["ns","B-ns","I-ns"]:
if i == 0:
adjust_line.append("B-ns")
elif adjust_line[i-1] in ["ns","B-ns","I-ns"]:
adjust_line.append("I-ns")
else:
adjust_line.append("B-ns")
elif label == "O":
adjust_line.append("O")
else:
if i == 0:
adjust_line.append(label)
elif adjust_line[i-1][-3:] == "-ns" and label[:2]=="I-":
adjust_line.append("B-"+label[-2:])
else:
adjust_line.append(label)
return adjust_line
def parse_token(line:List): # [for tokenf1]
"""
Given the predicted sequence (contain "ns" or "B-ns", "I-ns"), return the parsed BIO sequence (only contain "B-ns").
e.g.
Input : ns ns ns O B-playlist_owner B-playlist I-playlist O
Return : B-ns B-ns B-ns O B-playlist_owner B-playlist I-playlist O
Params:
- line(list),the prediced sequence (contain "ns" or "B-ns", "I-ns").
Returns:
- adjust_line(list), the parsed BIO sequence (contain "B-ns")
"""
adjust_line = []
for i,label in enumerate(line):
if label in ["ns","B-ns","I-ns"]:
adjust_line.append("B-ns")
elif label == "O":
adjust_line.append("O")
else:
if i == 0:
adjust_line.append(label)
elif line[i-1][-2:] == "ns" and label[:2]=="I-":
adjust_line.append("B-"+label[-2:])
else:
adjust_line.append(label)
return adjust_line
def token_metric(true:list,predict:list):
"""
Get token-level metrics.
"""
spanf1 = NSDSpanBasedF1Measure(tag_namespace="labels",
ignore_classes=[],
label_encoding="BIO",
nsd_slots=["ns"]
)
spanf1(pd.Series([true]),pd.Series([predict]),False)
token_metrics = spanf1.get_metric(reset=True)
f_nsd = token_metrics["f1-nsd"]
r_nsd = token_metrics["recall-nsd"]
p_nsd = token_metrics["precision-nsd"]
print(f"=====> Token(Experiment) <=====")
print(f"NSD: f:{f_nsd}, p:{p_nsd}, r:{r_nsd}\n")
def rose_metric(test_true_lines:list,test_pred_lines:list):
"""
To meet a reasonable NSD scenario, we propose a new metric, restriction-oriented span evaluation(ROSE).
We consider a span is correct:
- When the tokens prediction exceeds the span.
- When the number of correctly predicted tokens is greater than a settable proportion p of the span length.
Params:
- test_true_lines, (seq_dim,token_dim) contation BIO tags.
- test_pred_lines, (seq_dim,token_dim) contation BIO tags.
Returns:
-
"""
spanf1 = NSDSpanBasedF1Measure(tag_namespace="labels",
ignore_classes=[],
label_encoding="BIO",
nsd_slots=["ns"]
)
spanf1(pd.Series(test_true_lines),pd.Series(test_pred_lines),True,0.25)
nsd_metrics_25 = spanf1.get_metric(reset=True)
spanf1(pd.Series(test_true_lines),pd.Series(test_pred_lines),True,0.5)
nsd_metrics_50 = spanf1.get_metric(reset=True)
spanf1(pd.Series(test_true_lines),pd.Series(test_pred_lines),True,0.75)
nsd_metrics_75 = spanf1.get_metric(reset=True)
spanf1(pd.Series(test_true_lines),pd.Series(test_pred_lines),True,1)
nsd_metrics_100 = spanf1.get_metric(reset=True)
rose_f1_25 = nsd_metrics_25["f1-nsd"].round(2)
rose_f1_50 = nsd_metrics_50["f1-nsd"].round(2)
rose_f1_75 = nsd_metrics_75["f1-nsd"].round(2)
rose_f1_100 = nsd_metrics_100["f1-nsd"].round(2)
rose_p_25 = nsd_metrics_25["precision-nsd"].round(2)
rose_p_50 = nsd_metrics_50["precision-nsd"].round(2)
rose_p_75 = nsd_metrics_75["precision-nsd"].round(2)
rose_p_100 = nsd_metrics_100["precision-nsd"].round(2)
rose_r_25 = nsd_metrics_25["recall-nsd"].round(2)
rose_r_50 = nsd_metrics_50["recall-nsd"].round(2)
rose_r_75 = nsd_metrics_75["recall-nsd"].round(2)
rose_r_100 = nsd_metrics_100["recall-nsd"].round(2)
print(f"=====> ROSE(Experiment) <=====")
print(f"ROSE-25%: f:{rose_f1_25}, p:{rose_p_25}, r:{rose_r_25}")
print(f"ROSE-50%: f:{rose_f1_50}, p:{rose_p_50}, r:{rose_r_50}")
print(f"ROSE-75%: f:{rose_f1_75}, p:{rose_p_75}, r:{rose_r_75}")
print(f"ROSE-100%: f:{rose_f1_100}, p:{rose_p_100}, r:{rose_r_100}")