-
Notifications
You must be signed in to change notification settings - Fork 3
/
metric.py
317 lines (271 loc) · 9.28 KB
/
metric.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
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
""" Methods to compute classification statistics from the
prediction and labels.
"""
import numpy as np
from sklearn.metrics import confusion_matrix
from sklearn.metrics import accuracy_score
from sklearn.metrics import f1_score
from sklearn.metrics import matthews_corrcoef
def build_confusion_from_volume(true_label, pred_label):
"""Function to build the confusion matrix.
Parameters
----------
true_label: ndarray
Ground-truth array.
pred_label: ndarray
Prediction label given by the machine learning method.
Returns
-------
cm: ndarray
The resulting confusion matrix.
"""
return confusion_matrix(true_label, pred_label)
def labels_to_sensitivity_specificity(true_label, pred_label):
"""Function to compute the sensitivity and specificty.
Parameters
----------
true_label: ndarray
Ground-truth array.
pred_label: ndarray
Prediction label given by the machine learning method.
Returns
-------
sens, spec: double
The resulting sensitivity and specificity.
The sensitivity is also known as true positive rate, hit rate,
or recall. The specificity is also known as true negative rate.
"""
# Compute the confusion matrix
cm = build_confusion_from_volume(true_label, pred_label)
# Compute the sensitivity and specificity
if cm[1, 1] > 0:
sens = float(cm[1, 1]) / float(cm[1, 1] + cm[1, 0])
else:
sens = 0.
if cm[0, 0] > 0:
spec = float(cm[0, 0]) / float(cm[0, 0] + cm[0, 1])
else:
spec = 0.
return (sens, spec)
def labels_to_precision_negative_predictive_value(true_label, pred_label):
"""Function to compute the precision and the negative predictive value.
Parameters
----------
true_label: ndarray
Ground-truth array.
pred_label: ndarray
Prediction label given by the machine learning method.
Returns
-------
prec, npv: double
The resulting precision and negative predictive value.
The precision is also kown as the positive predictive value.
"""
# Compute the confusion matrix
cm = build_confusion_from_volume(true_label, pred_label)
# Compute the sensitivity and specificity
if cm[1, 1] > 0:
prec = float(cm[1, 1]) / float(cm[1, 1] + cm[0, 1])
else:
prec = 0.
if cm[0, 0] > 0:
npv = float(cm[0, 0]) / float(cm[0, 0] + cm[1, 0])
else:
npv = 0.
return (prec, npv)
def labels_to_geometric_mean(true_label, pred_label):
"""Function to compute the geometric mean.
Parameters
----------
true_label: ndarray
Ground-truth array.
pred_label: ndarray
Prediction label given by the machine learning method.
Returns
-------
gmean: double
The resulting geometric mean of the accuracies measured.
References
----------
.. [1] Kubat, M. and Matwin, S. "Addressing the curse of
imbalanced training sets: one-sided selection" ICML (1997)
"""
# Compute the confusion matrix
cm = build_confusion_from_volume(true_label, pred_label)
# Compute the sensitivity and specificity
sens = float(cm[1, 1]) / float(cm[1, 1] + cm[1, 0])
spec = float(cm[0, 0]) / float(cm[0, 0] + cm[0, 1])
# Compute the geometric mean
gmean = np.sqrt(sens * spec)
return gmean
def labels_to_accuracy(true_label, pred_label):
"""Function to compute the accuracy score.
Parameters
----------
true_label: ndarray
Ground-truth array.
pred_label: ndarray
Prediction label given by the machine learning method.
Returns
-------
acc: double
The resulting accuracy.
"""
return accuracy_score(true_label, pred_label)
def cost_with_bias(sens, spec, bias_pos, bias_neg):
"""Function to compute cost given the sensitivity and specificity.
----------
sens: float
Sensitivity.
spec: float
Specificity.
bias_pos: float
Constant influencing the bias of the positive class.
bias_neg: float
Constant influecing the bias of the negative class.
Returns
-------
cval: double
The resulting cost value.
"""
return (bias_pos * (1.0 - sens) +
bias_neg * (1.0 - spec)) / (bias_pos + bias_neg)
def labels_to_cost_value(true_label, pred_label, bias_pos=1.5, bias_neg=1.):
"""Function to compute cost given the ground-truth label
and predicted label.
----------
true_label: ndarray
Ground-truth array.
pred_label: ndarray
Prediction label given by the machine learning method.
bias_pos: float
Constant influencing the bias of the positive class.
bias_neg: float
Constant influecing the bias of the negative class.
Returns
-------
cval: double
The resulting cost value.
"""
sens, spec = labels_to_sensitivity_specificity(true_label, pred_label)
return cost_with_bias(sens, spec, bias_pos, bias_neg)
def labels_to_f1_score(true_label, pred_label):
"""Function to compute the F1 score.
Parameters
----------
true_label: ndarray
Ground-truth array.
pred_label: ndarray
Prediction label given by the machine learning method.
Returns
-------
f1score: double
The resulting F1 score.
"""
return f1_score(true_label, pred_label)
def labels_to_matthew_corrcoef(true_label, pred_label):
"""Function to compute the Matthew correlation coefficient.
Parameters
----------
true_label: ndarray
Ground-truth array.
pred_label: ndarray
Prediction label given by the machine learning method.
Returns
-------
mcc: double
The resulting Matthew correlation coefficient.
"""
return matthews_corrcoef(true_label, pred_label)
def labels_to_generalized_index_balanced_accuracy(true_label,
pred_label,
M='gmean',
alpha=0.1,
squared=True):
"""Function to compute the the generalized index of balanced accuracy.
Parameters
----------
true_label: ndarray
Ground-truth array.
pred_label: ndarray
Prediction label given by the machine learning method.
M: str (default: 'gmean')
Name of the metric to consider
``sens``:
use the sensitivity.
``'spec'``:
use the specificity.
``'prec'``:
use the precision.
``'npv'``:
use the negative predictive value.
``'gmean'``:
use the geometric mean.
``'acc'``:
use the accuracy score.
``'cost'``:
use the cost value.
``'f1score'``:
use the F1 score.
``'mcc'``:
use the Matthew correlation coefficient.
alpha: float (default: 0.1)
Dominance weight
squared: bool (default: True)
If the metric M should be squared
Returns
-------
iba: double
The resulting generalized index of balanced accuracy.
References
----------
.. [1] Garcia, V. and Mollineda, R.A. and Sanchez, J.S. "Theoretical
analysis of a performance measure for imbalanced data" ICPR (2010)
"""
if M == 'sens':
# Compute the sensitivity
met, _ = labels_to_sensitivity_specificity(true_label, pred_label)
elif M == 'spec':
# Compute the specificity
_, met = labels_to_sensitivity_specificity(true_label, pred_label)
elif M == 'prec':
# Compute the precision
met, _ = labels_to_precision_negative_predictive_value(true_label,
pred_label)
elif M == 'npv':
# Compute the precision
_, met = labels_to_precision_negative_predictive_value(true_label,
pred_label)
elif M == 'gmean':
# Compute the negative predictive value
met = labels_to_geometric_mean(true_label, pred_label)
elif M == 'acc':
# Compute the accuracy
met = labels_to_accuracy(true_label, pred_label)
elif M == 'cost':
# Compute the accuracy
met = labels_to_cost_value(true_label, pred_label)
elif M == 'f1score':
# Compute the F1 Score
met = labels_to_f1_score(true_label, pred_label)
elif M == 'mcc':
# Compute the Matthew correlation coefficient
met = labels_to_matthew_corrcoef(true_label, pred_label)
else:
raise ValueError('protoclass.metric.GIBA: The metric that you'
' attend to correct is not implemented.')
# Check the value of alpha is meaningful
if not ((alpha >= 0.) and (alpha <= 1.)):
raise ValueError('protoclass.metric.GIBA: The value of alpha'
' sould be set between 0 and 1.')
# Check if we should square the metric
if squared:
met = met ** 2
# Compute the dominance
# We need the sensitivity and specificity
sens, spec = labels_to_sensitivity_specificity(true_label, pred_label)
# Compute the dominance as the difference of the sensitiy and specificity
dom = float(sens - spec)
# Compute the generalized index of balanced accuracy
iba = (1. + float(alpha) * dom) * met
return iba