evaluate.py 18 KB
Newer Older
Manuel Günther's avatar
Manuel Günther committed
1
2
3
4
#!/usr/bin/env python
# vim: set fileencoding=utf-8 :

"""This script evaluates the given score files and computes EER, HTER.
5
6
7
It also is able to plot CMC and ROC curves.
You can set the environment variable BOB_NO_STYLE_CHANGES to any value to avoid
this script from changing the matplotlib style values. """
Manuel Günther's avatar
Manuel Günther committed
8

9
from __future__ import print_function
Manuel Günther's avatar
Manuel Günther committed
10
11
12
import bob.measure

import argparse
13
14
import numpy
import math
Manuel Günther's avatar
Manuel Günther committed
15
16
17
import os

# matplotlib stuff
18
19
import matplotlib
matplotlib.use('pdf')  # avoids TkInter threaded start
Manuel Günther's avatar
Manuel Günther committed
20
21
22
from matplotlib import pyplot
from matplotlib.backends.backend_pdf import PdfPages

23
24
25
26
27
28
if not os.environ.get('BOB_NO_STYLE_CHANGES'):
  # increase the default line width and font size
  matplotlib.rc('lines', linewidth=4)
  matplotlib.rc('font', size=18)


Manuel Günther's avatar
Manuel Günther committed
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
import bob.core
logger = bob.core.log.setup("bob.bio.base")


def command_line_arguments(command_line_parameters):
  """Parse the program options"""

  # set up command line parser
  parser = argparse.ArgumentParser(description=__doc__,
      formatter_class=argparse.ArgumentDefaultsHelpFormatter)

  parser.add_argument('-d', '--dev-files', required=True, nargs='+', help = "A list of score files of the development set.")
  parser.add_argument('-e', '--eval-files', nargs='+', help = "A list of score files of the evaluation set; if given it must be the same number of files as the --dev-files.")

  parser.add_argument('-s', '--directory', default = '.', help = "A directory, where to find the --dev-files and the --eval-files")

Manuel Günther's avatar
Manuel Günther committed
45
46
  parser.add_argument('-c', '--criterion', choices = ('EER', 'HTER', 'FAR'), help = "If given, the threshold of the development set will be computed with this criterion.")
  parser.add_argument('-f', '--far-value', type=float, default=0.1, help = "The FAR value in %% for which to evaluate (only for --criterion FAR)")
Manuel Günther's avatar
Manuel Günther committed
47
48
49
50
  parser.add_argument('-x', '--cllr', action = 'store_true', help = "If given, Cllr and minCllr will be computed.")
  parser.add_argument('-m', '--mindcf', action = 'store_true', help = "If given, minDCF will be computed.")
  parser.add_argument('--cost', default=0.99,  help='Cost for FAR in minDCF')
  parser.add_argument('-r', '--rr', action = 'store_true', help = "If given, the Recognition Rate will be computed.")
51
  parser.add_argument('-t', '--thresholds', type=float, nargs='+', help = "If given, the Recognition Rate will incorporate an Open Set handling, rejecting all scores that are below the given threshold; when multiple thresholds are given, they are applied in the same order as the --dev-files.")
Manuel Günther's avatar
Manuel Günther committed
52
53
54
  parser.add_argument('-l', '--legends', nargs='+', help = "A list of legend strings used for ROC, CMC and DET plots; if given, must be the same number than --dev-files.")
  parser.add_argument('-F', '--legend-font-size', type=int, default=18, help = "Set the font size of the legends.")
  parser.add_argument('-P', '--legend-position', type=int, help = "Set the font size of the legends.")
Manuel Günther's avatar
Manuel Günther committed
55
  parser.add_argument('-T', '--title', nargs = '+', help = "Overwrite the default title of the plot for development (and evaluation) set")
Manuel Günther's avatar
Manuel Günther committed
56
57
58
  parser.add_argument('-R', '--roc', help = "If given, ROC curves will be plotted into the given pdf file.")
  parser.add_argument('-D', '--det', help = "If given, DET curves will be plotted into the given pdf file.")
  parser.add_argument('-C', '--cmc', help = "If given, CMC curves will be plotted into the given pdf file.")
André Anjos's avatar
André Anjos committed
59
  parser.add_argument('-E', '--epc', help = "If given, EPC curves will be plotted into the given pdf file. For this plot --eval-files is mandatory.")
60
  parser.add_argument('--parser', default = '4column', choices = ('4column', '5column'), help="The style of the resulting score files. The default fits to the usual output of score files.")
Manuel Günther's avatar
Manuel Günther committed
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84

  # add verbose option
  bob.core.log.add_command_line_option(parser)

  # parse arguments
  args = parser.parse_args(command_line_parameters)

  # set verbosity level
  bob.core.log.set_verbosity_level(logger, args.verbose)


  # some sanity checks:
  if args.eval_files is not None and len(args.dev_files) != len(args.eval_files):
    logger.error("The number of --dev-files (%d) and --eval-files (%d) are not identical", len(args.dev_files), len(args.eval_files))

  # update legends when they are not specified on command line
  if args.legends is None:
    args.legends = [f.replace('_', '-') for f in args.dev_files]
    logger.warn("Legends are not specified; using legends estimated from --dev-files: %s", args.legends)

  # check that the legends have the same length as the dev-files
  if len(args.dev_files) != len(args.legends):
    logger.error("The number of --dev-files (%d) and --legends (%d) are not identical", len(args.dev_files), len(args.legends))

85
86
87
88
89
90
91
92
  if args.thresholds is not None:
    if len(args.thresholds) == 1:
      args.thresholds = args.thresholds * len(args.dev_files)
    elif len(args.thresholds) != len(args.dev_files):
      logger.error("If given, the number of --thresholds imust be either 1, or the same as --dev-files (%d), but it is %d", len(args.dev_files), len(args.thresholds))
  else:
    args.thresholds = [None] * len(args.dev_files)

Manuel Günther's avatar
Manuel Günther committed
93
94
95
96
97
98
  if args.title is not None:
    if args.eval_files is None and len(args.title) != 1:
      logger.warning("Ignoring the title for the evaluation set, as no evaluation set is given")
    if args.eval_files is not None and len(args.title) < 2:
      logger.error("The title for the evaluation set is not specified")

Manuel Günther's avatar
Manuel Günther committed
99
100
101
102
103
104
105
106
107
108
109
110
111
112
  return args


def _plot_roc(frrs, colors, labels, title, fontsize=18, position=None):
  if position is None: position = 4
  figure = pyplot.figure()
  # plot FAR and CAR for each algorithm
  for i in range(len(frrs)):
    pyplot.semilogx([100.0*f for f in frrs[i][0]], [100. - 100.0*f for f in frrs[i][1]], color=colors[i], lw=2, ms=10, mew=1.5, label=labels[i])

  # finalize plot
  pyplot.plot([0.1,0.1],[0,100], "--", color=(0.3,0.3,0.3))
  pyplot.axis([frrs[0][0][0]*100,100,0,100])
  pyplot.xticks((0.01, 0.1, 1, 10, 100), ('0.01', '0.1', '1', '10', '100'))
Manuel Günther's avatar
Manuel Günther committed
113
114
  pyplot.xlabel('FAR (%)')
  pyplot.ylabel('CAR (%)')
Manuel Günther's avatar
Manuel Günther committed
115
116
117
  pyplot.grid(True, color=(0.6,0.6,0.6))
  pyplot.legend(loc=position, prop = {'size':fontsize})
  pyplot.title(title)
Manuel Günther's avatar
Manuel Günther committed
118
  figure.set_tight_layout(True)
Manuel Günther's avatar
Manuel Günther committed
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139

  return figure


def _plot_det(dets, colors, labels, title, fontsize=18, position=None):
  if position is None: position = 1
  # open new page for current plot
  figure = pyplot.figure(figsize=(8.2,8))

  # plot the DET curves
  for i in range(len(dets)):
    pyplot.plot(dets[i][0], dets[i][1], color=colors[i], lw=2, ms=10, mew=1.5, label=labels[i])

  # change axes accordingly
  det_list = [0.0002, 0.001, 0.005, 0.01, 0.02, 0.05, 0.1, 0.2, 0.5, 0.7, 0.9, 0.95]
  ticks = [bob.measure.ppndf(d) for d in det_list]
  labels = [("%.5f" % (d*100)).rstrip('0').rstrip('.') for d in det_list]
  pyplot.xticks(ticks, labels)
  pyplot.yticks(ticks, labels)
  pyplot.axis((ticks[0], ticks[-1], ticks[0], ticks[-1]))

Manuel Günther's avatar
Manuel Günther committed
140
141
  pyplot.xlabel('FAR (%)')
  pyplot.ylabel('FRR (%)')
Manuel Günther's avatar
Manuel Günther committed
142
143
  pyplot.legend(loc=position, prop = {'size':fontsize})
  pyplot.title(title)
Manuel Günther's avatar
Manuel Günther committed
144
  figure.set_tight_layout(True)
Manuel Günther's avatar
Manuel Günther committed
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161

  return figure

def _plot_cmc(cmcs, colors, labels, title, fontsize=18, position=None):
  if position is None: position = 4
  # open new page for current plot
  figure = pyplot.figure()

  max_x = 0
  # plot the DET curves
  for i in range(len(cmcs)):
    x = bob.measure.plot.cmc(cmcs[i], figure=figure, color=colors[i], lw=2, ms=10, mew=1.5, label=labels[i])
    max_x = max(x, max_x)

  # change axes accordingly
  ticks = [int(t) for t in pyplot.xticks()[0]]
  pyplot.xlabel('Rank')
Manuel Günther's avatar
Manuel Günther committed
162
  pyplot.ylabel('Probability (%)')
Manuel Günther's avatar
Manuel Günther committed
163
164
165
166
  pyplot.xticks(ticks, [str(t) for t in ticks])
  pyplot.axis([0, max_x, 0, 100])
  pyplot.legend(loc=position, prop = {'size':fontsize})
  pyplot.title(title)
Manuel Günther's avatar
Manuel Günther committed
167
  figure.set_tight_layout(True)
Manuel Günther's avatar
Manuel Günther committed
168
169

  return figure
André Anjos's avatar
André Anjos committed
170
171
172



173
174
175
176
177
178
179
def _plot_epc(scores_dev, scores_eval, colors, labels, title, fontsize=18, position=None):
  if position is None: position = 4
  # open new page for current plot
  figure = pyplot.figure()

  # plot the DET curves
  for i in range(len(scores_dev)):
Manuel Günther's avatar
Manuel Günther committed
180
    bob.measure.plot.epc(scores_dev[i][0], scores_dev[i][1], scores_eval[i][0], scores_eval[i][1], 100, color=colors[i], label=labels[i], lw=2)
181
182
183

  # change axes accordingly
  pyplot.xlabel('alpha')
Manuel Günther's avatar
Manuel Günther committed
184
  pyplot.ylabel('HTER (%)')
185
186
187
188
  pyplot.title(title)
  pyplot.grid(True)
  pyplot.legend(loc=position, prop = {'size':fontsize})
  pyplot.title(title)
Manuel Günther's avatar
Manuel Günther committed
189
  figure.set_tight_layout(True)
190

André Anjos's avatar
André Anjos committed
191
  return figure
192

Manuel Günther's avatar
Manuel Günther committed
193
194
195
196
197
198
199
200
201
202
203


def main(command_line_parameters=None):
  """Reads score files, computes error measures and plots curves."""

  args = command_line_arguments(command_line_parameters)

  # get some colors for plotting
  cmap = pyplot.cm.get_cmap(name='hsv')
  colors = [cmap(i) for i in numpy.linspace(0, 1.0, len(args.dev_files)+1)]

Manuel Günther's avatar
Manuel Günther committed
204
  if args.criterion or args.roc or args.det or args.epc or args.cllr or args.mindcf:
Manuel Günther's avatar
Manuel Günther committed
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
    score_parser = {'4column' : bob.measure.load.split_four_column, '5column' : bob.measure.load.split_five_column}[args.parser]

    # First, read the score files
    logger.info("Loading %d score files of the development set", len(args.dev_files))
    scores_dev = [score_parser(os.path.join(args.directory, f)) for f in args.dev_files]

    if args.eval_files:
      logger.info("Loading %d score files of the evaluation set", len(args.eval_files))
      scores_eval = [score_parser(os.path.join(args.directory, f)) for f in args.eval_files]


    if args.criterion:
      logger.info("Computing %s on the development " % args.criterion + ("and HTER on the evaluation set" if args.eval_files else "set"))
      for i in range(len(scores_dev)):
        # compute threshold on development set
Manuel Günther's avatar
Manuel Günther committed
220
221
222
223
        if args.criterion == 'FAR':
          threshold = bob.measure.far_threshold(scores_dev[i][0], scores_dev[i][1], args.far_value/100.)
        else:
          threshold = {'EER': bob.measure.eer_threshold, 'HTER' : bob.measure.min_hter_threshold} [args.criterion](scores_dev[i][0], scores_dev[i][1])
Manuel Günther's avatar
Manuel Günther committed
224
225
        # apply threshold to development set
        far, frr = bob.measure.farfrr(scores_dev[i][0], scores_dev[i][1], threshold)
Manuel Günther's avatar
Manuel Günther committed
226
        if args.criterion == 'FAR':
André Anjos's avatar
André Anjos committed
227
228
          print("The FRR at FAR=%2.3f%% of the development set of '%s' is %2.3f%% (CAR: %2.3f%%)" % (args.far_value, args.legends[i], frr * 100., 100.*(1-frr)))
        else:
Manuel Günther's avatar
Manuel Günther committed
229
          print("The %s of the development set of '%s' is %2.3f%%" % (args.criterion, args.legends[i], (far + frr) * 50.)) # / 2 * 100%
Manuel Günther's avatar
Manuel Günther committed
230
231
232
        if args.eval_files:
          # apply threshold to evaluation set
          far, frr = bob.measure.farfrr(scores_eval[i][0], scores_eval[i][1], threshold)
Manuel Günther's avatar
Manuel Günther committed
233
234
          if args.criterion == 'FAR':
            print("The FRR of the evaluation set of '%s' is %2.3f%% (CAR: %2.3f%%)" % (args.legends[i], frr * 100., 100.*(1-frr))) # / 2 * 100%
André Anjos's avatar
André Anjos committed
235
          else:
Manuel Günther's avatar
Manuel Günther committed
236
            print("The HTER of the evaluation set of '%s' is %2.3f%%" % (args.legends[i], (far + frr) * 50.)) # / 2 * 100%
Manuel Günther's avatar
Manuel Günther committed
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


    if args.mindcf:
      logger.info("Computing minDCF on the development " + ("and on the evaluation set" if args.eval_files else "set"))
      for i in range(len(scores_dev)):
        # compute threshold on development set
        threshold = bob.measure.min_weighted_error_rate_threshold(scores_dev[i][0], scores_dev[i][1], args.cost)
        # apply threshold to development set
        far, frr = bob.measure.farfrr(scores_dev[i][0], scores_dev[i][1], threshold)
        print("The minDCF of the development set of '%s' is %2.3f%%" % (args.legends[i], (args.cost * far + (1-args.cost) * frr) * 100. ))
        if args.eval_files:
          # compute threshold on evaluation set
          threshold = bob.measure.min_weighted_error_rate_threshold(scores_eval[i][0], scores_eval[i][1], args.cost)
          # apply threshold to evaluation set
          far, frr = bob.measure.farfrr(scores_eval[i][0], scores_eval[i][1], threshold)
          print("The minDCF of the evaluation set of '%s' is %2.3f%%" % (args.legends[i], (args.cost * far + (1-args.cost) * frr) * 100. ))


    if args.cllr:
      logger.info("Computing Cllr and minCllr on the development " + ("and on the evaluation set" if args.eval_files else "set"))
      for i in range(len(scores_dev)):
        cllr = bob.measure.calibration.cllr(scores_dev[i][0], scores_dev[i][1])
        min_cllr = bob.measure.calibration.min_cllr(scores_dev[i][0], scores_dev[i][1])
        print("Calibration performance on development set of '%s' is Cllr %1.5f and minCllr %1.5f " % (args.legends[i], cllr, min_cllr))
        if args.eval_files:
          cllr = bob.measure.calibration.cllr(scores_eval[i][0], scores_eval[i][1])
          min_cllr = bob.measure.calibration.min_cllr(scores_eval[i][0], scores_eval[i][1])
          print("Calibration performance on evaluation set of '%s' is Cllr %1.5f and minCllr %1.5f" % (args.legends[i], cllr, min_cllr))


    if args.roc:
      logger.info("Computing CAR curves on the development " + ("and on the evaluation set" if args.eval_files else "set"))
      fars = [math.pow(10., i * 0.25) for i in range(-16,0)] + [1.]
      frrs_dev = [bob.measure.roc_for_far(scores[0], scores[1], fars) for scores in scores_dev]
      if args.eval_files:
        frrs_eval = [bob.measure.roc_for_far(scores[0], scores[1], fars) for scores in scores_eval]

      logger.info("Plotting ROC curves to file '%s'", args.roc)
      try:
        # create a multi-page PDF for the ROC curve
        pdf = PdfPages(args.roc)
        # create a separate figure for dev and eval
Manuel Günther's avatar
Manuel Günther committed
279
        pdf.savefig(_plot_roc(frrs_dev, colors, args.legends, args.title[0] if args.title is not None else "ROC curve for development set", args.legend_font_size, args.legend_position))
Manuel Günther's avatar
Manuel Günther committed
280
281
        del frrs_dev
        if args.eval_files:
Manuel Günther's avatar
Manuel Günther committed
282
          pdf.savefig(_plot_roc(frrs_eval, colors, args.legends, args.title[1] if args.title is not None else "ROC curve for evaluation set", args.legend_font_size, args.legend_position))
Manuel Günther's avatar
Manuel Günther committed
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
          del frrs_eval
        pdf.close()
      except RuntimeError as e:
        raise RuntimeError("During plotting of ROC curves, the following exception occured:\n%s\nUsually this happens when the label contains characters that LaTeX cannot parse." % e)

    if args.det:
      logger.info("Computing DET curves on the development " + ("and on the evaluation set" if args.eval_files else "set"))
      dets_dev = [bob.measure.det(scores[0], scores[1], 1000) for scores in scores_dev]
      if args.eval_files:
        dets_eval = [bob.measure.det(scores[0], scores[1], 1000) for scores in scores_eval]

      logger.info("Plotting DET curves to file '%s'", args.det)
      try:
        # create a multi-page PDF for the ROC curve
        pdf = PdfPages(args.det)
        # create a separate figure for dev and eval
Manuel Günther's avatar
Manuel Günther committed
299
        pdf.savefig(_plot_det(dets_dev, colors, args.legends, args.title[0] if args.title is not None else "DET plot for development set", args.legend_font_size, args.legend_position))
Manuel Günther's avatar
Manuel Günther committed
300
301
        del dets_dev
        if args.eval_files:
Manuel Günther's avatar
Manuel Günther committed
302
          pdf.savefig(_plot_det(dets_eval, colors, args.legends, args.title[1] if args.title is not None else "DET plot for evaluation set", args.legend_font_size, args.legend_position))
Manuel Günther's avatar
Manuel Günther committed
303
304
305
306
307
308
          del dets_eval
        pdf.close()
      except RuntimeError as e:
        raise RuntimeError("During plotting of ROC curves, the following exception occured:\n%s\nUsually this happens when the label contains characters that LaTeX cannot parse." % e)


309
    if args.epc:
Manuel Günther's avatar
Manuel Günther committed
310
      logger.info("Plotting EPC curves to file '%s'", args.epc)
André Anjos's avatar
André Anjos committed
311

312
313
      if not args.eval_files:
        raise ValueError("To plot the EPC curve the evaluation scores are necessary. Please, set it with the --eval-files option.")
André Anjos's avatar
André Anjos committed
314

315
316
317
      try:
        # create a multi-page PDF for the ROC curve
        pdf = PdfPages(args.epc)
Manuel Günther's avatar
Manuel Günther committed
318
        pdf.savefig(_plot_epc(scores_dev, scores_eval, colors, args.legends, args.title if args.title is not None else "EPC Curves" , args.legend_font_size, args.legend_position))
319
320
321
322
323
324
325
        pdf.close()
      except RuntimeError as e:
        raise RuntimeError("During plotting of EPC curves, the following exception occured:\n%s\nUsually this happens when the label contains characters that LaTeX cannot parse." % e)




Manuel Günther's avatar
Manuel Günther committed
326
327
328
329
330
331
332
  if args.cmc or args.rr:
    logger.info("Loading CMC data on the development " + ("and on the evaluation set" if args.eval_files else "set"))
    cmc_parser = {'4column' : bob.measure.load.cmc_four_column, '5column' : bob.measure.load.cmc_five_column}[args.parser]
    cmcs_dev = [cmc_parser(os.path.join(args.directory, f)) for f in args.dev_files]
    if args.eval_files:
      cmcs_eval = [cmc_parser(os.path.join(args.directory, f)) for f in args.eval_files]

333
334
335
336
337
338
    if args.cmc:
      logger.info("Plotting CMC curves to file '%s'", args.cmc)
      try:
        # create a multi-page PDF for the ROC curve
        pdf = PdfPages(args.cmc)
        # create a separate figure for dev and eval
Manuel Günther's avatar
Manuel Günther committed
339
        pdf.savefig(_plot_cmc(cmcs_dev, colors, args.legends, args.title[0] if args.title is not None else "CMC curve for development set", args.legend_font_size, args.legend_position))
340
        if args.eval_files:
Manuel Günther's avatar
Manuel Günther committed
341
          pdf.savefig(_plot_cmc(cmcs_eval, colors, args.legends, args.title[1] if args.title is not None else "CMC curve for evaluation set", args.legend_font_size, args.legend_position))
342
343
344
345
346
347
348
349
        pdf.close()
      except RuntimeError as e:
        raise RuntimeError("During plotting of ROC curves, the following exception occured:\n%s\nUsually this happens when the label contains characters that LaTeX cannot parse." % e)

    if args.rr:
      logger.info("Computing recognition rate on the development " + ("and on the evaluation set" if args.eval_files else "set"))
      for i in range(len(cmcs_dev)):
        rr = bob.measure.recognition_rate(cmcs_dev[i], args.thresholds[i])
Manuel Günther's avatar
Manuel Günther committed
350
        print("The Recognition Rate of the development set of '%s' is %2.3f%%" % (args.legends[i], rr * 100.))
351
352
353
        if args.eval_files:
          rr = bob.measure.recognition_rate(cmcs_eval[i], args.thresholds[i])
          print("The Recognition Rate of the development set of '%s' is %2.3f%%" % (args.legends[i], rr * 100.))