csv_dataset.py 14.7 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
#!/usr/bin/env python
# vim: set fileencoding=utf-8 :


import os
from bob.pipelines import Sample, DelayedSample, SampleSet
import csv
import bob.io.base
import functools
from abc import ABCMeta, abstractmethod
11
12
import numpy as np
import itertools
13
14


Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
15
class CSVBaseSampleLoader(metaclass=ABCMeta):
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
    """
    Convert CSV files in the format below to either a list of
    :any:`bob.pipelines.DelayedSample` or :any:`bob.pipelines.SampleSet`

    .. code-block:: text

       PATH,SUBJECT
       path_1,subject_1
       path_2,subject_2
       path_i,subject_j
       ...

    .. note::
       This class should be extended

    Parameters
    ----------

        data_loader:
            A python function that can be called parameterlessly, to load the
            sample in question from whatever medium

        extension:
            The file extension

    """

Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
43
    def __init__(self, data_loader, dataset_original_directory="", extension=""):
44
45
        self.data_loader = data_loader
        self.extension = extension
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
46
        self.dataset_original_directory = dataset_original_directory
47
48
49
50
51
52
53
54
55
56
57
        self.excluding_attributes = ["_data", "load", "key"]

    @abstractmethod
    def __call__(self, filename):
        pass

    @abstractmethod
    def convert_row_to_sample(self, row, header):
        pass

    @abstractmethod
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
58
59
60
    def convert_samples_to_samplesets(
        self, samples, group_by_subject=True, references=None
    ):
61
62
63
        pass


Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
64
class CSVToSampleLoader(CSVBaseSampleLoader):
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
    """
    Simple mechanism to convert CSV files in the format below to either a list of
    :any:`bob.pipelines.DelayedSample` or :any:`bob.pipelines.SampleSet`
    """

    def __call__(self, filename):
        def check_header(header):
            """
            A header should have at least "SUBJECT" AND "PATH"
            """
            header = [h.lower() for h in header]
            if not "subject" in header:
                raise ValueError(
                    "The field `subject` is not available in your dataset."
                )

            if not "path" in header:
                raise ValueError("The field `path` is not available in your dataset.")

        with open(filename) as cf:
            reader = csv.reader(cf)
            header = next(reader)

            check_header(header)
            return [self.convert_row_to_sample(row, header) for row in reader]

    def convert_row_to_sample(self, row, header):
        path = row[0]
        subject = row[1]
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
94
        kwargs = dict([[h, r] for h, r in zip(header[2:], row[2:])])
95
        return DelayedSample(
96
97
98
99
            functools.partial(
                self.data_loader,
                os.path.join(self.dataset_original_directory, path + self.extension),
            ),
100
101
102
103
104
            key=path,
            subject=subject,
            **kwargs,
        )

Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
105
106
107
    def convert_samples_to_samplesets(
        self, samples, group_by_subject=True, references=None
    ):
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
        def get_attribute_from_sample(sample):
            return dict(
                [
                    [attribute, sample.__dict__[attribute]]
                    for attribute in list(sample.__dict__.keys())
                    if attribute not in self.excluding_attributes
                ]
            )

        if group_by_subject:

            # Grouping sample sets
            sample_sets = dict()
            for s in samples:
                if s.subject not in sample_sets:
                    sample_sets[s.subject] = SampleSet(
                        [s], **get_attribute_from_sample(s)
                    )
126
127
                else:
                    sample_sets[s.subject].append(s)
128
129
130
            return list(sample_sets.values())

        else:
131
132
133
134
            return [
                SampleSet([s], **get_attribute_from_sample(s), references=references)
                for s in samples
            ]
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


class CSVDatasetDevEval:
    """
    Generic filelist dataset for :any:`bob.bio.base.pipelines.VanillaBiometrics` pipeline.
    Check :ref:`vanilla_biometrics_features` for more details about the Vanilla Biometrics Dataset
    interface.

    To create a new dataset, you need to provide a directory structure similar to the one below:

    .. code-block:: text

       my_dataset/
       my_dataset/my_protocol/
       my_dataset/my_protocol/train.csv
       my_dataset/my_protocol/train.csv/dev_enroll.csv
       my_dataset/my_protocol/train.csv/dev_probe.csv
       my_dataset/my_protocol/train.csv/eval_enroll.csv
       my_dataset/my_protocol/train.csv/eval_probe.csv
       ...


    In the above directory structure, inside of `my_dataset` should contain the directories with all
    evaluation protocols this dataset might have.
    Inside of the `my_protocol` directory should contain at least two csv files:

     - dev_enroll.csv
     - dev_probe.csv


    Those csv files should contain in each row i-) the path to raw data and ii-) the subject label
    for enrollment (:ref:`bob.bio.base.pipelines.vanilla_biometrics.abstract_classes.Database.references`) and
    probing (:ref:`bob.bio.base.pipelines.vanilla_biometrics.abstract_classes.Database.probes`).
    The structure of each CSV file should be as below:

    .. code-block:: text

       PATH,SUBJECT
       path_1,subject_1
       path_2,subject_2
       path_i,subject_j
       ...

    
    You might want to ship metadata within your Samples (e.g gender, age, annotation, ...)
    To do so is simple, just do as below:

    .. code-block:: text

       PATH,SUBJECT,METADATA_1,METADATA_2,METADATA_k
       path_1,subject_1,A,B,C
       path_2,subject_2,A,B,1
       path_i,subject_j,2,3,4
       ...


    The files `my_dataset/my_protocol/train.csv/eval_enroll.csv` and `my_dataset/my_protocol/train.csv/eval_probe.csv`
    are optional and it is used in case a protocol contains data for evaluation.
    
    Finally, the content of the file `my_dataset/my_protocol/train.csv` is used in the case a protocol
    contains data for training (:ref:`bob.bio.base.pipelines.vanilla_biometrics.abstract_classes.Database.background_model_samples`)

    Parameters
    ----------

        dataset_path: str
          Absolute path of the dataset protocol description

Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
203
        protocol_na,e: str
204
205
          The name of the protocol

206
207
208
        csv_to_sample_loader: :any:`CSVBaseSampleLoader`
            Base class that whose objective is to generate :any:`bob.pipelines.Samples`
            and/or :any:`bob.pipelines.SampleSet` from csv rows
209
210
211
212
213

    """

    def __init__(
        self,
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
214
215
        dataset_protocol_path,
        protocol_name,
216
        csv_to_sample_loader=CSVToSampleLoader(
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
217
            data_loader=bob.io.base.load, dataset_original_directory="", extension=""
218
219
220
221
        ),
    ):
        def get_paths():

Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
222
223
            if not os.path.exists(dataset_protocol_path):
                raise ValueError(f"The path `{dataset_protocol_path}` was not found")
224
225

            # TODO: Unzip file if dataset path is a zip
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
226
            protocol_path = os.path.join(dataset_protocol_path, protocol_name)
227
            if not os.path.exists(protocol_path):
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
228
                raise ValueError(f"The protocol `{protocol_name}` was not found")
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

            train_csv = os.path.join(protocol_path, "train.csv")
            dev_enroll_csv = os.path.join(protocol_path, "dev_enroll.csv")
            dev_probe_csv = os.path.join(protocol_path, "dev_probe.csv")
            eval_enroll_csv = os.path.join(protocol_path, "eval_enroll.csv")
            eval_probe_csv = os.path.join(protocol_path, "eval_probe.csv")

            # The minimum required is to have `dev_enroll_csv` and `dev_probe_csv`
            train_csv = train_csv if os.path.exists(train_csv) else None

            # Eval
            eval_enroll_csv = (
                eval_enroll_csv if os.path.exists(eval_enroll_csv) else None
            )
            eval_probe_csv = eval_probe_csv if os.path.exists(eval_probe_csv) else None

            # Dev
            if not os.path.exists(dev_enroll_csv):
                raise ValueError(
                    f"The file `{dev_enroll_csv}` is required and it was not found"
                )

            if not os.path.exists(dev_probe_csv):
                raise ValueError(
                    f"The file `{dev_probe_csv}` is required and it was not found"
                )

            return (
                train_csv,
                dev_enroll_csv,
                dev_probe_csv,
                eval_enroll_csv,
                eval_probe_csv,
            )

        (
            self.train_csv,
            self.dev_enroll_csv,
            self.dev_probe_csv,
            self.eval_enroll_csv,
            self.eval_probe_csv,
        ) = get_paths()

        def get_dict_cache():
            cache = dict()
            cache["train"] = None
            cache["dev_enroll_csv"] = None
            cache["dev_probe_csv"] = None
            cache["eval_enroll_csv"] = None
            cache["eval_probe_csv"] = None
            return cache

        self.cache = get_dict_cache()
        self.csv_to_sample_loader = csv_to_sample_loader

    def background_model_samples(self):

        self.cache["train"] = (
            self.csv_to_sample_loader(self.train_csv)
            if self.cache["train"] is None
            else self.cache["train"]
        )

        return self.cache["train"]

    def _get_samplesets(self, group="dev", purpose="enroll", group_by_subject=False):

Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
296
        fetching_probes = False
297
298
299
        if purpose == "enroll":
            cache_label = "dev_enroll_csv" if group == "dev" else "eval_enroll_csv"
        else:
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
300
            fetching_probes = True
301
302
303
304
305
            cache_label = "dev_probe_csv" if group == "dev" else "eval_probe_csv"

        if self.cache[cache_label] is not None:
            return self.cache[cache_label]

Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
306
307
        references = None
        if fetching_probes:
308
            references = list(set([s.subject for s in self.references(group=group)]))
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
309
310

        samples = self.csv_to_sample_loader(self.__dict__[cache_label])
311
312

        sample_sets = self.csv_to_sample_loader.convert_samples_to_samplesets(
Tiago de Freitas Pereira's avatar
Tiago de Freitas Pereira committed
313
            samples, group_by_subject=group_by_subject, references=references
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
        )

        self.cache[cache_label] = sample_sets

        return self.cache[cache_label]

    def references(self, group="dev"):
        return self._get_samplesets(
            group=group, purpose="enroll", group_by_subject=True
        )

    def probes(self, group="dev"):
        return self._get_samplesets(
            group=group, purpose="probe", group_by_subject=False
        )
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
465
466
467
468
469


class CSVDatasetCrossValidation:
    """
    Generic filelist dataset for :any:`bob.bio.base.pipelines.VanillaBiometrics` pipeline that 
    handles **CROSS VALIDATION**.

    Check :ref:`vanilla_biometrics_features` for more details about the Vanilla Biometrics Dataset
    interface.


    This interface will take one `csv_file` as input and split into i-) data for training and
    ii-) data for testing.
    The data for testing will be further split in data for enrollment and data for probing.
    The input CSV file should be casted in the following format:

    .. code-block:: text

       PATH,SUBJECT
       path_1,subject_1
       path_2,subject_2
       path_i,subject_j
       ...

    Parameters
    ----------

    csv_file_name: str
      CSV file containing all the samples from your database

    random_state: int
      Pseudo-random number generator seed

    test_size: float
      Percentage of the subjects used for testing

    samples_for_enrollment: float
      Number of samples used for enrollment

    csv_to_sample_loader: :any:`CSVBaseSampleLoader`
        Base class that whose objective is to generate :any:`bob.pipelines.Samples`
        and/or :any:`bob.pipelines.SampleSet` from csv rows

    """

    def __init__(
        self,
        csv_file_name="metadata.csv",
        random_state=0,
        test_size=0.8,
        samples_for_enrollment=1,
        csv_to_sample_loader=CSVToSampleLoader(
            data_loader=bob.io.base.load, dataset_original_directory="", extension=""
        ),
    ):
        def get_dict_cache():
            cache = dict()
            cache["train"] = None
            cache["dev_enroll_csv"] = None
            cache["dev_probe_csv"] = None
            return cache

        self.random_state = random_state
        self.cache = get_dict_cache()
        self.csv_to_sample_loader = csv_to_sample_loader
        self.csv_file_name = csv_file_name
        self.samples_for_enrollment = samples_for_enrollment
        self.test_size = test_size

        if self.test_size < 0 and self.test_size > 1:
            raise ValueError(
                f"`test_size` should be between 0 and 1. {test_size} is provided"
            )

    def _do_cross_validation(self):

        # Shuffling samples by subject
        samples_by_subject = group_samples_by_subject(
            self.csv_to_sample_loader(self.csv_file_name)
        )
        subjects = list(samples_by_subject.keys())
        np.random.seed(self.random_state)
        np.random.shuffle(subjects)

        # Getting the training data
        n_samples_for_training = len(subjects) - int(self.test_size * len(subjects))
        self.cache["train"] = list(
            itertools.chain(
                *[samples_by_subject[s] for s in subjects[0:n_samples_for_training]]
            )
        )

        # Splitting enroll and probe
        self.cache["dev_enroll_csv"] = []
        self.cache["dev_probe_csv"] = []
        for s in subjects[n_samples_for_training:]:
            samples = samples_by_subject[s]
            if len(samples) < self.samples_for_enrollment:
                raise ValueError(
                    f"Not enough samples ({len(samples)}) for enrollment for the subject {s}"
                )

            # Enrollment samples
            self.cache["dev_enroll_csv"].append(
                self.csv_to_sample_loader.convert_samples_to_samplesets(
                    samples[0 : self.samples_for_enrollment]
                )[0]
            )

            self.cache[
                "dev_probe_csv"
            ] += self.csv_to_sample_loader.convert_samples_to_samplesets(
                samples[self.samples_for_enrollment :],
                group_by_subject=False,
                references=subjects[n_samples_for_training:],
            )

    def _load_from_cache(self, cache_key):
        if self.cache[cache_key] is None:
            self._do_cross_validation()
        return self.cache[cache_key]

    def background_model_samples(self):
        return self._load_from_cache("train")

    def references(self, group="dev"):
        return self._load_from_cache("dev_enroll_csv")

    def probes(self, group="dev"):
        return self._load_from_cache("dev_probe_csv")


def group_samples_by_subject(samples):

    # Grouping sample sets
    samples_by_subject = dict()
    for s in samples:
        if s.subject not in samples_by_subject:
            samples_by_subject[s.subject] = []
        samples_by_subject[s.subject].append(s)
    return samples_by_subject