error_utils.py 17 KB
Newer Older
1
2
3
4
5
6
7
#!/usr/bin/env python
# Ivana Chingovska <ivana.chingovska@idiap.ch>
# Fri Dec  7 12:33:37 CET 2012
"""Utility functions for computation of EPSC curve and related measurement"""

import numpy

Amir MOHAMMADI's avatar
Amir MOHAMMADI committed
8
9
from bob.measure import farfrr

10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28

def calc_pass_rate(threshold, attacks):
    """Calculates the rate of successful spoofing attacks

    Parameters
    ----------
    threshold :
      the threshold used for classification
    scores :
      numpy with the scores of the spoofing attacks

    Returns
    -------
    float
      rate of successful spoofing attacks
    """
    return (attacks >= threshold).mean()


Amir MOHAMMADI's avatar
Amir MOHAMMADI committed
29
30
31
def weighted_neg_error_rate_criteria(
    data, weight, thres, beta=0.5, criteria="eer"
):
32
33
34
35
36
37
38
39
    """Given the single value for the weight parameter balancing between
    impostors and spoofing attacks and a threshold, calculates the error rates
    and their relationship depending on the criteria (difference in case of
    'eer', hter in case of 'min-hter' criteria)
    Keyword parameters:

      - data - the development data used to determine the threshold. List on 4
      numpy.arrays containing: negatives (licit), positives (licit),
40
      negatives (spoof), positives (spoof)
41
42
43
44
45
      - weight - the weight parameter balancing between impostors and spoofing
      attacks
      - thres - the given threshold
      - beta - the weight parameter balancing between real accesses and all the
      negative samples (impostors and spoofing attacks). Note that this
46
      parameter will be overridden and not considered if the selected criteria
47
48
      is 'min-hter'.
      - criteria - 'eer', 'wer' or 'min-hter' criteria for decision threshold
Yannick DAYER's avatar
Yannick DAYER committed
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

    licit_neg = data[0]
    licit_pos = data[1]
    spoof_neg = data[2]
    spoof_pos = data[3]  # unpacking the data
    farfrr_licit = farfrr(licit_neg, licit_pos, thres)
    farfrr_spoof = farfrr(spoof_neg, spoof_pos, thres)

    frr = farfrr_licit[1]  # farfrr_spoof[1] should have the same value
    far_i = farfrr_licit[0]
    far_s = farfrr_spoof[0]

    far_w = (1 - weight) * far_i + weight * far_s

    if criteria == "eer":
        if beta == 0.5:
            return abs(far_w - frr)
        else:
            # return abs(far_w - frr)
            return abs((1 - beta) * frr - beta * far_w)

    elif criteria == "min-hter":
        return (far_w + frr) / 2

    else:
        return (1 - beta) * frr + beta * far_w


def epsc_weights(licit_neg, licit_pos, spoof_neg, spoof_pos, points=100):
    """Returns the weights for EPSC

Yannick DAYER's avatar
Yannick DAYER committed
81
    Keyword arguments:
82

Yannick DAYER's avatar
Yannick DAYER committed
83
84
      - points - number of points to calculate EPSC
    """
85
86
87
88
89
    step_size = 1 / float(points)
    weights = numpy.array([(i * step_size) for i in range(points + 1)])
    return weights


Amir MOHAMMADI's avatar
Amir MOHAMMADI committed
90
91
92
def recursive_thr_search(
    data, span_min, span_max, weight, beta=0.5, criteria="eer"
):
93
    """Recursive search for the optimal threshold given a criteria. It
Yannick DAYER's avatar
Yannick DAYER committed
94
95
96
97
98
99
100
101
      evaluates the full range of thresholds at 100 points, and computes the one
      which optimizes the threshold. In the next search iteration, it examines
      the region around the point that optimizes the threshold. The procedure
      stops when the search range is smaller then 1e-10.

    Keyword arguments:
      - data - the development data used to determine the threshold. List on 4
      numpy.arrays containing: negatives (licit), positives (licit), negatives
102
      (spoof), positives (spoof)
Yannick DAYER's avatar
Yannick DAYER committed
103
104
105
106
107
108
109
110
111
112
113
      - span_min - the minimum of the search range
      - span_max - the maximum of the search range
      - weight - the weight parameter balancing between impostors and spoofing
      attacks
      - beta - the weight parameter balancing between real accesses and all the
      negative samples (impostors and spoofing attacks). Note that methods called
      within this function will override this parameter and not considered if the
      selected criteria is 'min-hter'.
      - criteria - the decision threshold criteria ('eer' for EER, 'wer' for
      Minimum WER or 'min-hter' for Minimum HTER criteria).
    """
114
115
116
117
118
119
120

    quit_thr = 1e-10
    steps = 100
    if abs((span_max - span_min) / span_max) < quit_thr:
        return span_max  # or span_min, it doesn't matter
    else:
        step_size = (span_max - span_min) / steps
Amir MOHAMMADI's avatar
Amir MOHAMMADI committed
121
122
123
        thresholds = numpy.array(
            [(i * step_size) + span_min for i in range(steps + 1)]
        )
124
125
        weighted_error_rates = numpy.array(
            [
Amir MOHAMMADI's avatar
Amir MOHAMMADI committed
126
127
128
                weighted_neg_error_rate_criteria(
                    data, weight, thr, beta, criteria
                )
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
                for thr in thresholds
            ]
        )
        selected_thres = thresholds[
            numpy.where(weighted_error_rates == min(weighted_error_rates))
        ]  # all the thresholds which have minimum weighted error rate
        thr = selected_thres[
            int(selected_thres.size / 2)
        ]  # choose the centrally positioned threshold
        return recursive_thr_search(
            data, thr - step_size, thr + step_size, weight, beta, criteria
        )


def weighted_negatives_threshold(
    licit_neg, licit_pos, spoof_neg, spoof_pos, weight, beta=0.5, criteria="eer"
):
    """Calculates the threshold for achieving the given criteria between the
Yannick DAYER's avatar
Yannick DAYER committed
147
148
149
150
      FAR_w and the FRR, given the single value for the weight parameter
      balancing between impostors and spoofing attacks and a single value for the
      parameter beta balancing between the real accesses and the negatives
      (impostors and spoofing attacks)
151

Yannick DAYER's avatar
Yannick DAYER committed
152
153
154
155
156
157
158
159
160
161
162
163
164
165
    Keyword parameters:
      - licit_neg - numpy.array of scores for the negatives (licit scenario)
      - licit_pos - numpy.array of scores for the positives (licit scenario)
      - spoof_neg - numpy.array of scores for the negatives (spoof scenario)
      - spoof_pos - numpy.array of scores for the positives (spoof scenario)
      - weight - the weight parameter balancing between impostors and spoofing
      attacks
      - beta - the weight parameter balancing between real accesses and all the
      negative samples (impostors and spoofing attacks). Note that methods called
      within this function will override this parameter and not considered if the
      selected criteria is 'min-hter'.
      - criteria - the decision threshold criteria ('eer' for EER, 'wer' for
      Minimum WER or 'min-hter' for Minimum HTER criteria).
    """
166
167
168
169
170
171
172
173
174
175
176
177
    span_min = min(
        numpy.append(licit_neg, spoof_neg)
    )  # the min of the span where we will search for the threshold
    span_max = max(
        numpy.append(licit_pos, spoof_pos)
    )  # the max of the span where we will search for the threshold
    data = (
        licit_neg,
        licit_pos,
        spoof_neg,
        spoof_pos,
    )  # pack the data into a single list
Amir MOHAMMADI's avatar
Amir MOHAMMADI committed
178
179
180
    return recursive_thr_search(
        data, span_min, span_max, weight, beta, criteria
    )
181
182
183
184
185
186
187
188
189
190
191
192
193


def epsc_thresholds(
    licit_neg,
    licit_pos,
    spoof_neg,
    spoof_pos,
    points=100,
    criteria="eer",
    omega=None,
    beta=None,
):
    """Calculates the optimal thresholds for EPSC, for a range of the weight
Yannick DAYER's avatar
Yannick DAYER committed
194
195
196
      parameter balancing between impostors and spoofing attacks, and for a range
      of the beta parameter balancing between real accesses and all the negatives
      (impostors and spoofing attacks)
197

Yannick DAYER's avatar
Yannick DAYER committed
198
    Keyword arguments:
199

Yannick DAYER's avatar
Yannick DAYER committed
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
      - licit_neg - numpy.array of scores for the negatives (licit scenario)
      - licit_pos - numpy.array of scores for the positives (licit scenario)
      - spoof_neg - numpy.array of scores for the negatives (spoof scenario)
      - spoof_pos - numpy.array of scores for the positives (spoof scenario)
      - points - number of points to calculate EPSC
      - criteria - the decision threshold criteria ('eer', 'wer' or 'min-hter')
      - omega - the value of the parameter omega, balancing between impostors and
      spoofing attacks. If None, it is going to span the full range [0,1].
      Otherwise, can be set to a fixed value or a list of values.
      - beta - the value of the parameter beta, balancing between real accesses
      and all the negatives (zero-effort impostors and spoofing attacks). If
      None, it is going to span the full range [0,1]. Otherwise, can be set to a
      fixed value or a list of values.

    """
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
    step_size = 1 / float(points)

    if omega is None:
        omega = numpy.array([(i * step_size) for i in range(points + 1)])
    elif (
        not isinstance(omega, list)
        and not isinstance(omega, tuple)
        and not isinstance(omega, numpy.ndarray)
    ):
        omega = numpy.array([omega])
    else:
        omega = numpy.array(omega)

    if beta is None:
        beta = numpy.array([(i * step_size) for i in range(points + 1)])
    elif (
        not isinstance(beta, list)
        and not isinstance(beta, tuple)
        and not isinstance(beta, numpy.ndarray)
    ):
        beta = numpy.array([beta])
    else:
        beta = numpy.array(beta)

    thresholds = numpy.ndarray([beta.size, omega.size], "float64")
    for bindex, b in enumerate(beta):
        thresholds[bindex, :] = numpy.array(
            [
                weighted_negatives_threshold(
Amir MOHAMMADI's avatar
Amir MOHAMMADI committed
244
245
246
247
248
249
250
                    licit_neg,
                    licit_pos,
                    spoof_neg,
                    spoof_pos,
                    w,
                    b,
                    criteria=criteria,
251
252
253
254
255
256
257
258
259
260
261
262
                )
                for w in omega
            ],
            "float64",
        )

    return omega, beta, thresholds


def weighted_err(error_1, error_2, weight):
    """Calculates the weighted error rate between the two input parameters

Yannick DAYER's avatar
Yannick DAYER committed
263
264
265
266
267
268
    Keyword arguments:
      - error_1 - the first input error rate (FAR for zero effort impostors
      usually)
      - error_2 - the second input error rate (SFAR)
      - weight - the given weight
    """
269
270
271
272
273
274
275
    return (1 - weight) * error_1 + weight * error_2


def error_rates_at_weight(
    licit_neg, licit_pos, spoof_neg, spoof_pos, omega, threshold, beta=0.5
):
    """Calculates several error rates: FRR, FAR (zero-effort impostors), SFAR,
Yannick DAYER's avatar
Yannick DAYER committed
276
277
        FAR_w, HTER_w for a given value of w. It returns the calculated threshold
        as a last argument
278

Yannick DAYER's avatar
Yannick DAYER committed
279
      Keyword arguments:
280

Yannick DAYER's avatar
Yannick DAYER committed
281
282
283
284
285
286
287
288
        - licit_neg - numpy.array of scores for the negatives (licit scenario)
        - licit_pos - numpy.array of scores for the positives (licit scenario)
        - spoof_neg - numpy.array of scores for the negatives (spoof scenario)
        - spoof_pos - numpy.array of scores for the positives (spoof scenario)
        - threshold - the given threshold
        - omega - the omega parameter balancing between impostors and spoofing
        attacks
        - beta - the weight parameter balancing between real accesses and all the
289

Yannick DAYER's avatar
Yannick DAYER committed
290
291
    negative samples (impostors and spoofing attacks).
    """
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
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464

    farfrr_licit = farfrr(
        licit_neg, licit_pos, threshold
    )  # calculate test frr @ threshold (licit scenario)
    farfrr_spoof = farfrr(
        spoof_neg, spoof_pos, threshold
    )  # calculate test frr @ threshold (spoof scenario)

    # we can take this value from farfrr_spoof as well, it doesn't matter
    frr = farfrr_licit[1]
    far = farfrr_licit[0]
    sfar = farfrr_spoof[0]

    far_w = weighted_err(far, sfar, omega)
    hter_w = (far_w + frr) / 2
    wer_wb = weighted_err(frr, far_w, beta)

    return (frr, far, sfar, far_w, wer_wb, hter_w, threshold)


def epsc_error_rates(
    licit_neg, licit_pos, spoof_neg, spoof_pos, thresholds, omega, beta
):
    """Calculates several error rates: FAR_w and WER_wb for the given weights
    (omega and beta) and thresholds (the thresholds need to be computed first
    using the method: epsc_thresholds() before passing to this method)

    Parameters
    ----------
    licit_neg : array_like
        array of scores for the negatives (licit scenario)
    licit_pos : array_like
        array of scores for the positives (licit scenario)
    spoof_neg : array_like
        array of scores for the negatives (spoof scenario)
    spoof_pos : array_like
        array of scores for the positives (spoof scenario)
    thresholds : array_like
        ndarray with threshold values
    omega : array_like
        array of the omega parameter balancing between impostors
        and spoofing attacks
    beta : array_like
        array of the beta parameter balancing between real accesses
        and all negatives (impostors and spoofing attacks)

    Returns
    -------
    far_w_errors: array_like
        FAR_w
    wer_wb_errors: array_like
        WER_wb
    """

    far_w_errors = numpy.ndarray((beta.size, omega.size), "float64")
    wer_wb_errors = numpy.ndarray((beta.size, omega.size), "float64")

    for bindex, b in enumerate(beta):
        errors = [
            error_rates_at_weight(
                licit_neg,
                licit_pos,
                spoof_neg,
                spoof_pos,
                w,
                thresholds[bindex, windex],
                b,
            )
            for windex, w in enumerate(omega)
        ]
        far_w_errors[bindex, :] = [errors[i][3] for i in range(len(errors))]
        wer_wb_errors[bindex, :] = [errors[i][4] for i in range(len(errors))]

    return far_w_errors, wer_wb_errors


def all_error_rates(
    licit_neg, licit_pos, spoof_neg, spoof_pos, thresholds, omega, beta
):
    """Calculates several error rates: FAR_w and WER_wb for the given weights
    (omega and beta) and thresholds (the thresholds need to be computed first
    using the method: epsc_thresholds() before passing to this method)

    Parameters
    ----------
    licit_neg : array_like
        array of scores for the negatives (licit scenario)
    licit_pos : array_like
        array of scores for the positives (licit scenario)
    spoof_neg : array_like
        array of scores for the negatives (spoof scenario)
    spoof_pos : array_like
        array of scores for the positives (spoof scenario)
    thresholds : array_like
        ndarray with threshold values
    omega : array_like
        array of the omega parameter balancing between impostors
        and spoofing attacks
    beta : array_like
        array of the beta parameter balancing between real accesses
        and all negatives (impostors and spoofing attacks)

    Returns
    -------
    far_w_errors: array_like
        FAR_w
    wer_wb_errors: array_like
        WER_wb
    """

    frr_errors = numpy.ndarray((beta.size, omega.size), "float64")
    far_errors = numpy.ndarray((beta.size, omega.size), "float64")
    sfar_errors = numpy.ndarray((beta.size, omega.size), "float64")
    far_w_errors = numpy.ndarray((beta.size, omega.size), "float64")
    wer_wb_errors = numpy.ndarray((beta.size, omega.size), "float64")
    hter_wb_errors = numpy.ndarray((beta.size, omega.size), "float64")

    for bindex, b in enumerate(beta):
        errors = [
            error_rates_at_weight(
                licit_neg,
                licit_pos,
                spoof_neg,
                spoof_pos,
                w,
                thresholds[bindex, windex],
                b,
            )
            for windex, w in enumerate(omega)
        ]
        frr_errors[bindex, :] = [errors[i][0] for i in range(len(errors))]
        far_errors[bindex, :] = [errors[i][1] for i in range(len(errors))]
        sfar_errors[bindex, :] = [errors[i][2] for i in range(len(errors))]
        far_w_errors[bindex, :] = [errors[i][3] for i in range(len(errors))]
        wer_wb_errors[bindex, :] = [errors[i][4] for i in range(len(errors))]
        hter_wb_errors[bindex, :] = [errors[i][5] for i in range(len(errors))]

    return (
        frr_errors,
        far_errors,
        sfar_errors,
        far_w_errors,
        wer_wb_errors,
        hter_wb_errors,
    )


def calc_aue(
    licit_neg,
    licit_pos,
    spoof_neg,
    spoof_pos,
    thresholds,
    omega,
    beta,
    l_bound=0,
    h_bound=1,
    var_param="omega",
):
    """Calculates AUE of EPSC for the given thresholds and weights

    Keyword arguments:

    - licit_neg - numpy.array of scores for the negatives (licit scenario)
    - licit_pos - numpy.array of scores for the positives (licit scenario)
    - spoof_neg - numpy.array of scores for the negatives (spoof scenario)
    - spoof_pos - numpy.array of scores for the positives (spoof scenario)
    - l_bound - lower bound of integration
    - h_bound - higher bound of integration
    - points - number of points to calculate EPSC
    - criteria - the decision threshold criteria ('eer', 'wer' or 'min-hter')
    - var_param - name of the parameter which is varied on the abscissa
    ('omega' or 'beta')
Yannick DAYER's avatar
Yannick DAYER committed
465
    """
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484

    from scipy import integrate

    if var_param == "omega":
        errors = all_error_rates(
            licit_neg, licit_pos, spoof_neg, spoof_pos, thresholds, omega, beta
        )
        weights = omega  # setting the weights to the varying parameter
    else:
        errors = all_error_rates(
            licit_neg, licit_pos, spoof_neg, spoof_pos, thresholds, omega, beta
        )
        weights = beta  # setting the weights to the varying parameter

    wer_errors = errors[4].reshape(1, errors[4].size)

    l_ind = numpy.where(weights >= l_bound)[0][0]
    h_ind = numpy.where(weights <= h_bound)[0][-1]
    aue = integrate.cumtrapz(wer_errors, weights)
Amir MOHAMMADI's avatar
Amir MOHAMMADI committed
485
486
487
    aue = numpy.append(
        [0], aue
    )  # for indexing purposes, aue is cumulative integration
488
489
490
    aue = aue[h_ind] - aue[l_ind]

    return aue