Skip to content

Model keras

ModelKeras

Bases: ModelClass

Generic model for Keras NN

Source code in template_vision/models_training/model_keras.py
 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
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
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
class ModelKeras(ModelClass):
    '''Generic model for Keras NN'''

    _default_name = 'model_keras'
    # Not implemented :
    # -> _get_model
    # -> reload_from_standalone

    # Should pby be overridden :
    # -> _get_preprocess_input

    def __init__(self, batch_size: int = 64, epochs: int = 99, validation_split: float = 0.2, patience: int = 5,
                 width: int = 224, height: int = 224, depth: int = 3, color_mode: str = 'rgb',
                 in_memory: bool = False, data_augmentation_params: dict = {},
                 nb_train_generator_images_to_save: int = 20,
                 keras_params: dict = {}, **kwargs) -> None:
        '''Initialization of the class (see ModelClass for more arguments)

        Kwargs:
            batch_size (int): Batch size
            epochs (int): Number of epochs
            validation_split (float): Percentage for the validation set split
                Only used if no input validation set when fitting
            patience (int): Early stopping patience
            width (int): NN input width (images are resized)
            height (int): NN input height (images are resized)
            depth (int): NN input depth
            color_mode (str): NN input color mode
            in_memory (bool): If all images should be loaded in memory, otherwise it uses a generator
                /!\\ OOM errors can happen really quickly (depends on the dataset size)
                /!\\ Data augmentation impossible if `in_memory` is set to True
            data_augmentation_params (dict): Dictionnary of parameters to be used with the data augmentation
                cf. https://www.tensorflow.org/api_docs/python/tf/keras/preprocessing/image/ImageDataGenerator
                /!\\ Not used if `in_memory` is set to True
            nb_train_generator_images_to_save (int): If > 0, save some input generated images
                If helps with to understand what goes in your NN
            keras_params (dict): Parameters used by Keras models.
                e.g. learning_rate, nb_lstm_units, etc...
                The purpose of this dictionary is for the user to use it as they wants in the _get_model function
                This parameter was initially added in order to do an hyperparameters search
        Raises:
            ValueError: If `in_memory` is set to True and `data_augmentation_params` is not empty
        '''
        # TODO: learning rate should be an attribute !

        # Check for errors
        if in_memory and len(data_augmentation_params) > 0:
            raise ValueError("Data augmentation is impossible for 'in_memory' mode")

        # Init.
        super().__init__(**kwargs)

        # Fix tensorflow GPU
        gpu_devices = tf.config.experimental.list_physical_devices('GPU')
        for device in gpu_devices:
            tf.config.experimental.set_memory_growth(device, True)

        # Get logger (must be done after super init)
        self.logger = logging.getLogger(__name__)

        # Param. model
        self.batch_size = batch_size
        self.epochs = epochs
        self.validation_split = validation_split
        self.patience = patience

        # Params. generator
        self.width = width
        self.height = height
        self.depth = depth
        self.color_mode = color_mode
        self.in_memory = in_memory
        self.data_augmentation_params = data_augmentation_params.copy()

        # Warnings if depth does not match with color_mode
        if self.color_mode == 'rgb' and self.depth != 3:
            self.logger.warning(f"`color_mode` parameter is 'rgb', but `depth` parameteris not equal to 3 ({self.depth})")
            self.logger.warning("We continue, but this can lead to errors during the training")
        if self.color_mode == 'rgba' and self.depth != 4:
            self.logger.warning(f"`color_mode` parameter is 'rgba', but `depth` parameteris not equal to 4 ({self.depth})")
            self.logger.warning("We continue, but this can lead to errors during the training")

        # TODO: add Test time augmentation ?

        # Misc.
        self.nb_train_generator_images_to_save = nb_train_generator_images_to_save

        # Model set on fit
        self.model: Any = None

        # Set preprocess input
        self.preprocess_input = self._get_preprocess_input()

        # Keras params
        self.keras_params = keras_params.copy()

        # Keras custom objects : we get the ones specified in utils_deep_keras
        self.custom_objects = utils_deep_keras.custom_objects

    def fit(self, df_train: pd.DataFrame, df_valid: Union[pd.DataFrame, None] = None, with_shuffle: bool = True, **kwargs) -> dict:
        '''Fits the model

        Args:
            df_train (pd.DataFrame): Train dataset
                Must contain file_path & file_class columns if classifier
                Must contain file_path & bboxes columns if object detector
        Kwargs:
            df_valid (pd.DataFrame): Validation dataset
                Must contain file_path & file_class columns if classifier
                Must contain file_path & bboxes columns if object detector
            with_shuffle (boolean): If the train dataset must be shuffled
                This should be used if the input dataset is not shuffled & no validation set as the split_validation takes the lines in order.
                Thus, the validation set might get classes which are not in the train set ...
        Raises:
            NotImplementedError: If the model is not `classifier` nor `object_detector`
        Returns:
            dict: Fit arguments, to be used with transfer learning fine-tuning
        '''
        if self.model_type == 'classifier':
            return self._fit_classifier(df_train, df_valid=df_valid, with_shuffle=with_shuffle, **kwargs)
        elif self.model_type == 'object_detector':
            return self._fit_object_detector(df_train, df_valid=df_valid, with_shuffle=with_shuffle, **kwargs)
        else:
            raise NotImplementedError("Only `classifier` and `object_detector` model type are supported.")

    def _fit_classifier(self, df_train: pd.DataFrame, df_valid: pd.DataFrame = None, with_shuffle: bool = True, **kwargs) -> dict:
        '''Fits the model - classifier

        Args:
            df_train (pd.DataFrame): Train dataset
                Must contain file_path & file_class columns
        Kwargs:
            df_valid (pd.DataFrame): Validation dataset
                Must contain file_path & file_class columns
            with_shuffle (boolean): If the train dataset must be shuffled
                This should be used if the input dataset is not shuffled & no validation set as the split_validation takes the lines in order.
                Thus, the validation set might get classes which are not in the train set ...
        Raises:
            ValueError: If the model is not of type `classifier`
            ValueError: If already trained and new dataset does not match model's classes
        Returns:
            dict: Fit arguments, to be used with transfer learning fine-tuning
        '''
        if self.model_type != 'classifier':
            raise ValueError(f"`_fit_classifier` function does not support model type {self.model_type}")

        ##############################################
        # Manage retrain
        ##############################################

        # If a model has already been fitted, we make a new folder in order not to overwrite the existing one !
        # And we save the old conf
        if self.trained:
            # Get src files to save
            src_files = [os.path.join(self.model_dir, "configurations.json")]
            if self.nb_fit > 1:
                for i in range(1, self.nb_fit):
                    src_files.append(os.path.join(self.model_dir, f"configurations_fit_{i}.json"))
            # Change model dir
            self.model_dir = self._get_new_model_dir()
            # Get dst files
            dst_files = [os.path.join(self.model_dir, f"configurations_fit_{self.nb_fit}.json")]
            if self.nb_fit > 1:
                for i in range(1, self.nb_fit):
                    dst_files.append(os.path.join(self.model_dir, f"configurations_fit_{i}.json"))
            # Copies
            for src, dst in zip(src_files, dst_files):
                try:
                    shutil.copyfile(src, dst)
                except Exception as e:
                    self.logger.error(f"Impossible to copy {src} to {dst}")
                    self.logger.error("We still continue ...")
                    self.logger.error(repr(e))

        ##############################################
        # Prepare dataset
        # Also extract list of classes
        ##############################################

        # Extract list of classes from df_train
        list_classes = sorted(list(df_train['file_class'].unique()))
        # Also set dict_classes
        dict_classes = {i: col for i, col in enumerate(list_classes)}

        # Validate classes if already trained, else set them
        if self.trained:
            if self.list_classes != list_classes:
                raise ValueError("Error: the new dataset does not match with the already fitted model")
            if self.dict_classes != dict_classes:
                raise ValueError("Error: the new dataset does not match with the already fitted model")
        else:
            self.list_classes = list_classes
            self.dict_classes = dict_classes

        # Shuffle training dataset if wanted
        # It is advised as validation_split from keras does not shufle the data
        # Hence, for classificationt task, we might have classes in the validation data that we never met in the training data
        if with_shuffle:
            df_train = df_train.sample(frac=1.).reset_index(drop=True)

        if df_valid is None:
            self.logger.warning(f"Warning, no validation set. The training set will be splitted (validation fraction = {self.validation_split})")

        ##############################################
        # We save some preprocessed / augmented input images examples
        ##############################################

        # Save some examples
        if self.nb_train_generator_images_to_save > 0:
            self.logger.info("Retrieving a generator to save some preprocessed / augmented input images examples")
            # 1. Retrieve a generator (if in_memory, use 'valid' to avoid augmentation)
            if not self.in_memory:
                tmp_gen = self._get_generator(df_train, data_type='train', batch_size=1)
            else:
                tmp_gen = self._get_generator(df_train, data_type='valid', batch_size=1)
            # 2. Retrieve generated images one by one
            images = [tmp_gen.next()[0][0] for i in range(self.nb_train_generator_images_to_save)]
            # 3. Remove negative pixels
            min_pixel = min([np.min(_) for _ in images])
            if min_pixel < 0:
                images = [arr - min_pixel for arr in images]
            # 4. Rescale and scale uint8
            max_pixel = max([np.max(_) for _ in images])
            images = [(arr * 255 / max_pixel).astype('uint8') for arr in images]
            # 5. Cast back to image format
            images = [Image.fromarray(arr, 'RGBA' if arr.shape[-1] == 4 else 'RGB') for arr in images]
            # 6. Save
            save_dir = os.path.join(self.model_dir, f'examples_fit_{self.nb_fit + 1}')
            if not os.path.exists(save_dir):
                os.makedirs(save_dir)
            for i, im in enumerate(images):
                im_path = os.path.join(save_dir, f'example_{i}.png')
                im.save(im_path, format='PNG')

        ##############################################
        # Get generators if not in_memory, else get full data
        # Finally fit the model
        ##############################################

        if not self.in_memory:
            self.logger.info("Loading data via generators")

            # Create generators
            if df_valid is not None:
                self.logger.info("Retrieving a generator for the training set")
                train_generator = self._get_generator(df_train, data_type='train', batch_size=min(self.batch_size, len(df_train)))
                self.logger.info("Retrieving a generator for the validation set")
                valid_generator = self._get_generator(df_valid, data_type='valid', batch_size=min(self.batch_size, len(df_valid)))
                # Set dataset related args
                steps_per_epoch_arg = len(df_train) // min(self.batch_size, len(df_train))
                validation_steps_arg = len(df_valid) // min(self.batch_size, len(df_valid))
            else:
                # If no validation, we'll split the training set using validation_split attribute
                df_train_split, df_valid_split = train_test_split(df_train, test_size=self.validation_split)
                self.logger.info("Retrieving a generator for the training set")
                train_generator = self._get_generator(df_train_split, data_type='train', batch_size=min(self.batch_size, len(df_train_split)))
                self.logger.info("Retrieving a generator for the validation set")
                valid_generator = self._get_generator(df_valid_split, data_type='valid', batch_size=min(self.batch_size, len(df_valid_split)))
                # Set dataset related args
                steps_per_epoch_arg = len(df_train_split) // min(self.batch_size, len(df_train_split))
                validation_steps_arg = len(df_valid_split) // min(self.batch_size, len(df_valid_split))

            # Get fit arguments
            x_arg = train_generator
            y_arg = None
            batch_size_arg = None
            # validation_data does work with generators (TensorFlow doc is not up to date)
            validation_data_arg = valid_generator
            validation_split_arg = None

        # Load in memory - Can easily lead to OOM issues
        else:
            self.logger.info("Loading data in memory")

            # We retrieve all the data
            # Trick: we still use generators to have the correct preprocessing
            # -> data_type = valid (no shuffle, no data augmentation)
            train_generator = self._get_generator(df_train, data_type='valid', batch_size=len(df_train))
            x_train, y_train = train_generator.next()
            if df_valid is not None:
                valid_generator = self._get_generator(df_valid, data_type='valid', batch_size=min(self.batch_size, len(df_valid)))
                x_val, y_val = valid_generator.next()
                validation_data = (x_val, y_val)
            else:
                validation_data = None

            # Get fit arguments
            x_arg = x_train
            y_arg = y_train
            batch_size_arg = self.batch_size
            steps_per_epoch_arg = None
            validation_data_arg = validation_data  # Can be None if no validation set
            validation_steps_arg = None
            validation_split_arg = self.validation_split if validation_data is None else None

        # Get model (if already fitted, _get_model returns instance model)
        self.model = self._get_model()

        # Get callbacks (early stopping & checkpoint)
        callbacks = self._get_callbacks()

        # Fit
        # We use a try...except in order to save the model if an error arises
        # after more than a minute into training
        start_time = time.time()
        try:
            fit_arguments = {
                'x': x_arg,
                'y': y_arg,
                'batch_size': batch_size_arg,
                'steps_per_epoch': steps_per_epoch_arg,
                'validation_data': validation_data_arg,
                'validation_split': validation_split_arg,
                'validation_steps': validation_steps_arg,
            }
            fit_history = self.model.fit(  # type: ignore
                epochs=self.epochs,
                callbacks=callbacks,
                verbose=1,
                **fit_arguments,
            )
        except (RuntimeError, SystemError, SystemExit, EnvironmentError, KeyboardInterrupt, tf.errors.ResourceExhaustedError, tf.errors.InternalError,
                tf.errors.UnavailableError, tf.errors.UnimplementedError, tf.errors.UnknownError, Exception) as e:
            # Steps:
            # 1. Display tensorflow error
            # 2. Check if more than one minute elapsed & not several iterations & existence best.hdf5
            # 3. Reload best model
            # 4. We consider that a fit occured (trained = True, nb_fit += 1)
            # 5. Save & create a warning file
            # 6. Display error messages
            # 7. Raise an error

            # 1.
            self.logger.error(repr(e))

            # 2.
            best_path = os.path.join(self.model_dir, 'best.hdf5')
            time_spent = time.time() - start_time
            if time_spent >= 60 and os.path.exists(best_path):
                # 3.
                self.model = load_model(best_path, custom_objects=self.custom_objects)
                # 4.
                self.trained = True
                self.nb_fit += 1
                # 5.
                self.save()
                with open(os.path.join(self.model_dir, "0_MODEL_INCOMPLETE"), 'w'):
                    pass
                with open(os.path.join(self.model_dir, "1_TRAINING_NEEDS_TO_BE_RESUMED"), 'w'):
                    pass
                # 6.
                self.logger.error("[EXPERIMENTAL] Error during model training")
                self.logger.error(f"[EXPERIMENTAL] The error happened after {round(time_spent, 2)}s of training")
                self.logger.error("[EXPERIMENTAL] A saving of the model is done but this model won't be usable as is.")
                self.logger.error(f"[EXPERIMENTAL] In order to resume the training, we have to specify this model ({ntpath.basename(self.model_dir)}) in the file 2_training.py")
                self.logger.error("[EXPERIMENTAL] Warning, the preprocessing is not saved in the configuration file")
                self.logger.error("[EXPERIMENTAL] Warning, the best model might be corrupted in some cases")
            # 7.
            raise RuntimeError("Error during model training")

        # Print accuracy & loss if level_save > 'LOW'
        if self.level_save in ['MEDIUM', 'HIGH']:
            self._plot_metrics_and_loss(fit_history)
            # Reload best model
            self.model = load_model(
                os.path.join(self.model_dir, 'best.hdf5'),
                custom_objects=self.custom_objects
            )

        # Set trained
        self.trained = True
        self.nb_fit += 1

        # Return fit arguments. This is useful for transfer learning algorithms
        return fit_arguments

    def _fit_object_detector(self, df_train: pd.DataFrame, df_valid: pd.DataFrame = None, with_shuffle: bool = True, **kwargs) -> dict:
        '''Fits the model - object detector

        Args:
            df_train (pd.DataFrame): Train dataset
                Must contain file_path & bboxes columns
        Kwargs:
            df_valid (pd.DataFrame): Validation dataset
                Must contain file_path & bboxes columns
            with_shuffle (boolean): If the train dataset must be shuffled
        Raises:
            ValueError: If the model is not of type `object_detector`
        '''
        raise NotImplementedError("'_fit_object_detector' needs to be overridden")

    @utils.trained_needed
    def predict(self, df_test: pd.DataFrame, return_proba: bool = False, batch_size: Union[int, None] = None) -> Union[np.ndarray, list]:
        '''Predictions on test set

        Args:
            df_test (pd.DataFrame): DataFrame to be predicted, with column file_path
        Kwargs:
            return_proba (bool): If the function should return the probabilities instead of the classes -- classifier only
            batch_size (int): Batch size to be used -- classifier only
        Raises:
            NotImplementedError: If the model is not `classifier` nor `object_detector`
        Returns:
            (np.ndarray | list): Array, shape = [n_samples, n_classes] or List of n_samples elements
        '''
        if self.model_type == 'classifier':
            return self._predict_classifier(df_test, return_proba=return_proba, batch_size=batch_size)
        elif self.model_type == 'object_detector':
            return self._predict_object_detector(df_test)
        else:
            raise NotImplementedError("Only 'classifier' and 'object_detector' model type are supported")

    @utils.trained_needed
    def _predict_classifier(self, df_test, return_proba: bool = False, batch_size: int = None) -> np.ndarray:
        '''Predictions on test set

        Args:
            df_test (pd.DataFrame): DataFrame to be predicted, with column file_path
        Kwargs:
            return_proba (bool): If the function should return the probabilities instead of the classes
            batch_size (int): Batch size to be used
        Raises:
            ValueError: If the model is not a classifier
        Returns:
            (np.ndarray): Array, shape = [n_samples, n_classes]
        '''
        if self.model_type != 'classifier':
            raise ValueError(f"`_predict_classifier` function does not support model type {self.model_type}")

        # Backup on training batch size if no batch size defined
        if batch_size is None:
            batch_size = self.batch_size

        # Get generator or fulldata if in_memory
        if not self.in_memory:
            self.logger.info("Retrieving a generator for test data")
            test_generator = self._get_generator(df_test, data_type='test', batch_size=min(batch_size, len(df_test)))
            # Get predict arguments
            x_arg = test_generator
            batch_size_arg = None
        else:
            self.logger.info("Retrieving a all test data in memory")
            test_generator = self._get_generator(df_test, data_type='test', batch_size=len(df_test))
            x_test, _ = test_generator.next()
            # Get predict arguments
            x_arg = x_test
            batch_size_arg = batch_size

        # Predict
        predicted_proba = self.model.predict(  # type: ignore
            x_arg,
            batch_size=batch_size_arg,
            steps=None,
            workers=8,  # TODO : Check if this is ok if there are less CPUs
            verbose=1
        )

        # We return the probabilities if wanted
        if return_proba:
            return predicted_proba

        # Finally, we get the classes predictions
        return self.get_classes_from_proba(predicted_proba)  # type: ignore

    @utils.trained_needed
    def _predict_object_detector(self, df_test: pd.DataFrame, **kwargs) -> list:
        '''Predictions on test set - works only with batch size = 1

        Args:
            df_test (pd.DataFrame): DataFrame to be predicted, with column file_path
        Raises:
            ValueError: If the model is not an object detector
        Returns:
            list: List of list of bboxes (one list per image)
        '''
        raise NotImplementedError("'_predict_object_detector' needs to be overridden")

    @utils.trained_needed
    def predict_proba(self, df_test, batch_size: int = None) -> np.ndarray:
        '''Predicts probabilities on the test dataset

        Args:
            df_test (pd.DataFrame): DataFrame to be predicted, with column file_path
        Kwargs:
            batch_size (int): Batch size to be used
        Raises:
            ValueError: If the model is not a classifier
        Returns:
            (np.ndarray): Array, shape = [n_samples, n_classes]
        '''
        if self.model_type != 'classifier':
            raise ValueError(f"`predict_proba` function does not support model type {self.model_type}")

        # We reuse the predict function
        return self.predict(df_test, return_proba=True, batch_size=batch_size)

    def _get_generator(self, df: pd.DataFrame, data_type: str, batch_size: int, **kwargs) -> ImageDataGenerator:
        '''Gets image generator from a list of files

        Args:
            df (pd.DataFrame): DataFrame with files to be loaded
            data_type (str): 'train', 'valid' or 'test'
            batch_size (int): Batch size to be used
        Raises:
            NotImplementedError: If the model type is not supported
        '''
        if self.model_type == 'classifier':
            return self._get_generator_classifier(df, data_type, batch_size)
        else:
            raise NotImplementedError(f"`_get_generator` needs to be overridden for model type {self.model_type}")

    def _get_generator_classifier(self, df: pd.DataFrame, data_type: str, batch_size: int, **kwarg) -> ImageDataGenerator:
        '''Gets image generator from a list of files - classifier version

        Args:
            df (pd.DataFrame): DataFrame with files to be loaded
            data_type (str): 'train', 'valid' or 'test'
            batch_size (int): Batch size to be used
        Raises:
            ValueError: If the model is not a classifier
            ValueError: If data_type is not in ['train', 'valid', 'test']
            AttributeError: If list_classes attribute is not defined
        '''
        if self.model_type != 'classifier':
            raise ValueError(f"`_get_generator_classifier` function does not support model type {self.model_type}")
        if data_type not in ['train', 'valid', 'test']:
            raise ValueError(f"{data_type} is not a valid option for argument data_type (['train', 'valid', 'test'])")
        if self.list_classes is None:
            raise AttributeError("Cannot get an image generator if list_classes is not set.")

        # Copy
        df = df.copy(deep=True)
        # Set data_gen (no augmentation if validation/test)
        if data_type == 'train':
            data_generator = ImageDataGenerator(preprocessing_function=self.preprocess_input, **self.data_augmentation_params)
        else:
            data_generator = ImageDataGenerator(preprocessing_function=self.preprocess_input)

        # Get generator
        shuffle = True if data_type == 'train' else False  # DO NOT SHUFFLE IF VALID OR TEST !
        if data_type != 'test':
            generator = data_generator.flow_from_dataframe(df, directory=None, x_col='file_path', y_col='file_class', classes=self.list_classes,
                                                           target_size=(self.width, self.height), color_mode=self.color_mode, class_mode='categorical',
                                                           batch_size=batch_size, shuffle=shuffle)
        # For the test dataset, we create a fake DataFrame with a unique class
        else:
            df['fake_class_col'] = 'all_classes'
            generator = data_generator.flow_from_dataframe(df, directory=None, x_col='file_path', y_col='fake_class_col', classes=['all_classes'],
                                                           target_size=(self.width, self.height), color_mode=self.color_mode, class_mode='categorical',
                                                           batch_size=batch_size, shuffle=False)

        return generator

    def _get_preprocess_input(self) -> Union[Callable, None]:
        '''Gets the preprocessing to be used before feeding images to the NN
        Needs to be overridden by child classes

        Returns:
            (Callable | None): Preprocessing function
        '''
        return None

    def _get_model(self) -> Any:
        '''Gets a model structure - returns the instance model instead if already defined

        Returns:
            (Model): a Keras model
        '''
        raise NotImplementedError("'_get_model' needs to be overridden")

    def _get_callbacks(self, *args) -> list:
        '''Gets model callbacks

        Returns:
            list: List of callbacks
        '''
        # Get classic callbacks
        callbacks = [EarlyStopping(monitor='val_loss', patience=self.patience, restore_best_weights=True)]
        if self.level_save in ['MEDIUM', 'HIGH']:
            callbacks.append(
                ModelCheckpoint(
                    filepath=os.path.join(self.model_dir, 'best.hdf5'), monitor='val_loss', save_best_only=True, mode='auto'
                )
            )
        callbacks.append(CSVLogger(filename=os.path.join(self.model_dir, 'logger.csv'), separator=';', append=False))
        callbacks.append(TerminateOnNaN())

        # Get LearningRateScheduler
        scheduler = self._get_learning_rate_scheduler()
        if scheduler is not None:
            callbacks.append(LearningRateScheduler(scheduler))

        # Manage tensorboard
        if self.level_save in ['HIGH']:
            # Get log directory
            models_path = utils.get_models_path()
            tensorboard_dir = os.path.join(models_path, 'tensorboard_logs')
            # We add a prefix so that the function load_model works correctly (it looks for a sub-folder with model name)
            log_dir = os.path.join(tensorboard_dir, f"tensorboard_{ntpath.basename(self.model_dir)}")
            if not os.path.exists(log_dir):
                os.makedirs(log_dir)

            # TODO: check if this class does not slow proccesses
            # -> For now: comment
            # Create custom class to monitore LR changes
            # https://stackoverflow.com/questions/49127214/keras-how-to-output-learning-rate-onto-tensorboard
            # class LRTensorBoard(TensorBoard):
            #     def __init__(self, log_dir, **kwargs) -> None:  # add other arguments to __init__ if you need
            #         super().__init__(log_dir=log_dir, **kwargs)
            #
            #     def on_epoch_end(self, epoch, logs=None):
            #         logs.update({'lr': K.eval(self.model.optimizer.lr)})
            #         super().on_epoch_end(epoch, logs)

            callbacks.append(TensorBoard(log_dir=log_dir, write_grads=False, write_images=False))
            self.logger.info(f"To start tensorboard: python -m tensorboard.main --logdir {tensorboard_dir} --samples_per_plugin images=10")
            # We use samples_per_plugin to avoid a rare issue between matplotlib and tensorboard
            # https://stackoverflow.com/questions/27147300/matplotlib-tcl-asyncdelete-async-handler-deleted-by-the-wrong-thread

        return callbacks

    def _get_learning_rate_scheduler(self) -> Union[Callable, None]:
        '''Fonction to define a Learning Rate Scheduler
           -> if it returns None, no scheduler will be used. (def.)
           -> This function will be save directly in the model configuration file
           -> This can be overridden at runing time

        Returns:
            (Callable | None): A learning rate Scheduler
        '''
        # e.g.
        # def scheduler(epoch):
        #     lim_epoch = 75
        #     if epoch < lim_epoch:
        #         return 0.01
        #     else:
        #         return max(0.001, 0.01 * math.exp(0.01 * (lim_epoch - epoch)))
        scheduler = None
        return scheduler

    def _plot_metrics_and_loss(self, fit_history, **kwargs) -> None:
        '''Plots available metrics and losses

        Args:
            fit_history (?) : fit history
        '''
        # Manage dir
        plots_path = os.path.join(self.model_dir, 'plots')
        if not os.path.exists(plots_path):
            os.makedirs(plots_path)

        # Get a dictionnary of possible metrics/loss plots
        metrics_dir = {
            'acc': ['Accuracy', 'accuracy'],
            'loss': ['Loss', 'loss'],
            'categorical_accuracy': ['Categorical accuracy', 'categorical_accuracy'],
            'f1': ['F1-score', 'f1_score'],
            'precision': ['Precision', 'precision'],
            'recall': ['Recall', 'recall'],
        }

        # Plot each available metric
        for metric in fit_history.history.keys():
            if metric in metrics_dir.keys():
                title = metrics_dir[metric][0]
                filename = metrics_dir[metric][1]
                plt.figure(figsize=(10, 8))
                plt.plot(fit_history.history[metric])
                plt.plot(fit_history.history[f'val_{metric}'])
                plt.title(f"Model {title}")
                plt.ylabel(title)
                plt.xlabel('Epoch')
                plt.legend(['Train', 'Validation'], loc='upper left')
                # Save
                filename = f"{filename}.jpeg"
                plt.savefig(os.path.join(plots_path, filename))

                # Close figures
                plt.close('all')

    def _save_model_png(self, model) -> None:
        '''Tries to save the structure of the model in png format
        Graphviz necessary

        Args:
            model (?): model to plot
        '''
        # Check if graphiz is intalled
        # TODO : to be improved !
        graphiz_path = 'C:/Program Files (x86)/Graphviz2.38/bin/'
        if os.path.isdir(graphiz_path):
            os.environ["PATH"] += os.pathsep + 'C:/Program Files (x86)/Graphviz2.38/bin/'
            img_path = os.path.join(self.model_dir, 'model.png')
            plot_model(model, to_file=img_path)

    @no_type_check  # We do not check the type, because it is complicated with managing custom_objects_str
    def save(self, json_data: Union[dict, None] = None) -> None:
        '''Saves the model

        Kwargs:
            json_data (dict): Additional configurations to be saved
        '''
        # Save configuration JSON
        if json_data is None:
            json_data = {}

        json_data['librairie'] = 'keras'
        json_data['batch_size'] = self.batch_size
        json_data['epochs'] = self.epochs
        json_data['validation_split'] = self.validation_split
        json_data['patience'] = self.patience
        json_data['width'] = self.width
        json_data['height'] = self.height
        json_data['depth'] = self.depth
        json_data['color_mode'] = self.color_mode
        json_data['in_memory'] = self.in_memory
        json_data['data_augmentation_params'] = self.data_augmentation_params
        json_data['nb_train_generator_images_to_save'] = self.nb_train_generator_images_to_save
        json_data['keras_params'] = self.keras_params
        if self.model is not None:
            json_data['keras_model'] = json.loads(self.model.to_json())
        else:
            json_data['keras_model'] = None

        # Add _get_model code if not in json_data
        if '_get_model' not in json_data.keys():
            json_data['_get_model'] = pickle.source.getsourcelines(self._get_model)[0]
        # Add _get_preprocess_input code if not in json_data
        if '_get_preprocess_input' not in json_data.keys():
            json_data['_get_preprocess_input'] = pickle.source.getsourcelines(self._get_preprocess_input)[0]
        # Save preprocess_input to a .pkl file if level_save > LOW
        pkl_path = os.path.join(self.model_dir, "preprocess_input.pkl")
        if self.level_save in ['MEDIUM', 'HIGH']:
            with open(pkl_path, 'wb') as f:
                pickle.dump(self.preprocess_input, f)
        # Add _get_learning_rate_scheduler code if not in json_data
        if '_get_learning_rate_scheduler' not in json_data.keys():
            json_data['_get_learning_rate_scheduler'] = pickle.source.getsourcelines(self._get_learning_rate_scheduler)[0]
        # Add custom_objects code if not in json_data
        if 'custom_objects' not in json_data.keys():
            custom_objects_str = self.custom_objects.copy()
            for key in custom_objects_str.keys():
                if callable(custom_objects_str[key]):
                    # Nominal case
                    if not type(custom_objects_str[key]) == functools.partial:
                        custom_objects_str[key] = pickle.source.getsourcelines(custom_objects_str[key])[0]
                    # Manage partials
                    else:
                        custom_objects_str[key] = {
                            'type': 'partial',
                            'args': custom_objects_str[key].args,
                            'function': pickle.source.getsourcelines(custom_objects_str[key].func)[0],
                        }
            json_data['custom_objects'] = custom_objects_str

        # Save strategy :
        # - best.hdf5 already saved in fit()
        # - can't pickle keras model, so we drop it, save, and reload it
        keras_model = self.model
        self.model = None
        super().save(json_data=json_data)
        self.model = keras_model

    def reload_model(self, hdf5_path: str) -> Any:
        '''Loads a Keras model from a HDF5 file

        Args:
            hdf5_path (str): Path to the hdf5 file
        Returns:
            ?: Keras model
        '''
        # Fix tensorflow GPU if not already done (useful if we reload a model)
        try:
            gpu_devices = tf.config.experimental.list_physical_devices('GPU')
            for device in gpu_devices:
                tf.config.experimental.set_memory_growth(device, True)
        except Exception:
            pass

        # We check if we already have the custom objects
        if hasattr(self, 'custom_objects') and self.custom_objects is not None:
            custom_objects = self.custom_objects
        else:
            self.logger.warning("Can't find the attribute 'custom_objects' in the model to be reloaded")
            self.logger.warning("Backup on the default custom_objects of utils_deep_keras")
            custom_objects = utils_deep_keras.custom_objects

        # Loading of the model
        keras_model = load_model(hdf5_path, custom_objects=custom_objects)

        # Set trained to true if not already true
        if not self.trained:
            self.trained = True
            self.nb_fit = 1

        # Return
        return keras_model

    def reload_from_standalone(self, **kwargs) -> None:
        '''Reloads a model from its configuration and "standalones" files
        - /!\\ Needs to be overridden /!\\ -
        '''
        raise NotImplementedError("'reload' needs to be overridden")

    def _is_gpu_activated(self) -> bool:
        '''Checks if a GPU is used

        Returns:
            bool: whether GPU is available or not
        '''
        # Check for available GPU devices
        physical_devices = tf.config.list_physical_devices('GPU')
        if len(physical_devices) > 0:
            return True
        else:
            return False

__init__(batch_size=64, epochs=99, validation_split=0.2, patience=5, width=224, height=224, depth=3, color_mode='rgb', in_memory=False, data_augmentation_params={}, nb_train_generator_images_to_save=20, keras_params={}, **kwargs)

Initialization of the class (see ModelClass for more arguments)

Kwargs

batch_size (int): Batch size epochs (int): Number of epochs validation_split (float): Percentage for the validation set split Only used if no input validation set when fitting patience (int): Early stopping patience width (int): NN input width (images are resized) height (int): NN input height (images are resized) depth (int): NN input depth color_mode (str): NN input color mode in_memory (bool): If all images should be loaded in memory, otherwise it uses a generator /! OOM errors can happen really quickly (depends on the dataset size) /! Data augmentation impossible if in_memory is set to True data_augmentation_params (dict): Dictionnary of parameters to be used with the data augmentation cf. https://www.tensorflow.org/api_docs/python/tf/keras/preprocessing/image/ImageDataGenerator /! Not used if in_memory is set to True nb_train_generator_images_to_save (int): If > 0, save some input generated images If helps with to understand what goes in your NN keras_params (dict): Parameters used by Keras models. e.g. learning_rate, nb_lstm_units, etc... The purpose of this dictionary is for the user to use it as they wants in the _get_model function This parameter was initially added in order to do an hyperparameters search

Raises: ValueError: If in_memory is set to True and data_augmentation_params is not empty

Source code in template_vision/models_training/model_keras.py
def __init__(self, batch_size: int = 64, epochs: int = 99, validation_split: float = 0.2, patience: int = 5,
             width: int = 224, height: int = 224, depth: int = 3, color_mode: str = 'rgb',
             in_memory: bool = False, data_augmentation_params: dict = {},
             nb_train_generator_images_to_save: int = 20,
             keras_params: dict = {}, **kwargs) -> None:
    '''Initialization of the class (see ModelClass for more arguments)

    Kwargs:
        batch_size (int): Batch size
        epochs (int): Number of epochs
        validation_split (float): Percentage for the validation set split
            Only used if no input validation set when fitting
        patience (int): Early stopping patience
        width (int): NN input width (images are resized)
        height (int): NN input height (images are resized)
        depth (int): NN input depth
        color_mode (str): NN input color mode
        in_memory (bool): If all images should be loaded in memory, otherwise it uses a generator
            /!\\ OOM errors can happen really quickly (depends on the dataset size)
            /!\\ Data augmentation impossible if `in_memory` is set to True
        data_augmentation_params (dict): Dictionnary of parameters to be used with the data augmentation
            cf. https://www.tensorflow.org/api_docs/python/tf/keras/preprocessing/image/ImageDataGenerator
            /!\\ Not used if `in_memory` is set to True
        nb_train_generator_images_to_save (int): If > 0, save some input generated images
            If helps with to understand what goes in your NN
        keras_params (dict): Parameters used by Keras models.
            e.g. learning_rate, nb_lstm_units, etc...
            The purpose of this dictionary is for the user to use it as they wants in the _get_model function
            This parameter was initially added in order to do an hyperparameters search
    Raises:
        ValueError: If `in_memory` is set to True and `data_augmentation_params` is not empty
    '''
    # TODO: learning rate should be an attribute !

    # Check for errors
    if in_memory and len(data_augmentation_params) > 0:
        raise ValueError("Data augmentation is impossible for 'in_memory' mode")

    # Init.
    super().__init__(**kwargs)

    # Fix tensorflow GPU
    gpu_devices = tf.config.experimental.list_physical_devices('GPU')
    for device in gpu_devices:
        tf.config.experimental.set_memory_growth(device, True)

    # Get logger (must be done after super init)
    self.logger = logging.getLogger(__name__)

    # Param. model
    self.batch_size = batch_size
    self.epochs = epochs
    self.validation_split = validation_split
    self.patience = patience

    # Params. generator
    self.width = width
    self.height = height
    self.depth = depth
    self.color_mode = color_mode
    self.in_memory = in_memory
    self.data_augmentation_params = data_augmentation_params.copy()

    # Warnings if depth does not match with color_mode
    if self.color_mode == 'rgb' and self.depth != 3:
        self.logger.warning(f"`color_mode` parameter is 'rgb', but `depth` parameteris not equal to 3 ({self.depth})")
        self.logger.warning("We continue, but this can lead to errors during the training")
    if self.color_mode == 'rgba' and self.depth != 4:
        self.logger.warning(f"`color_mode` parameter is 'rgba', but `depth` parameteris not equal to 4 ({self.depth})")
        self.logger.warning("We continue, but this can lead to errors during the training")

    # TODO: add Test time augmentation ?

    # Misc.
    self.nb_train_generator_images_to_save = nb_train_generator_images_to_save

    # Model set on fit
    self.model: Any = None

    # Set preprocess input
    self.preprocess_input = self._get_preprocess_input()

    # Keras params
    self.keras_params = keras_params.copy()

    # Keras custom objects : we get the ones specified in utils_deep_keras
    self.custom_objects = utils_deep_keras.custom_objects

fit(df_train, df_valid=None, with_shuffle=True, **kwargs)

Fits the model

Parameters:

Name Type Description Default
df_train DataFrame

Train dataset Must contain file_path & file_class columns if classifier Must contain file_path & bboxes columns if object detector

required

Kwargs: df_valid (pd.DataFrame): Validation dataset Must contain file_path & file_class columns if classifier Must contain file_path & bboxes columns if object detector with_shuffle (boolean): If the train dataset must be shuffled This should be used if the input dataset is not shuffled & no validation set as the split_validation takes the lines in order. Thus, the validation set might get classes which are not in the train set ... Raises: NotImplementedError: If the model is not classifier nor object_detector Returns: dict: Fit arguments, to be used with transfer learning fine-tuning

Source code in template_vision/models_training/model_keras.py
def fit(self, df_train: pd.DataFrame, df_valid: Union[pd.DataFrame, None] = None, with_shuffle: bool = True, **kwargs) -> dict:
    '''Fits the model

    Args:
        df_train (pd.DataFrame): Train dataset
            Must contain file_path & file_class columns if classifier
            Must contain file_path & bboxes columns if object detector
    Kwargs:
        df_valid (pd.DataFrame): Validation dataset
            Must contain file_path & file_class columns if classifier
            Must contain file_path & bboxes columns if object detector
        with_shuffle (boolean): If the train dataset must be shuffled
            This should be used if the input dataset is not shuffled & no validation set as the split_validation takes the lines in order.
            Thus, the validation set might get classes which are not in the train set ...
    Raises:
        NotImplementedError: If the model is not `classifier` nor `object_detector`
    Returns:
        dict: Fit arguments, to be used with transfer learning fine-tuning
    '''
    if self.model_type == 'classifier':
        return self._fit_classifier(df_train, df_valid=df_valid, with_shuffle=with_shuffle, **kwargs)
    elif self.model_type == 'object_detector':
        return self._fit_object_detector(df_train, df_valid=df_valid, with_shuffle=with_shuffle, **kwargs)
    else:
        raise NotImplementedError("Only `classifier` and `object_detector` model type are supported.")

predict(df_test, return_proba=False, batch_size=None)

Predictions on test set

Parameters:

Name Type Description Default
df_test DataFrame

DataFrame to be predicted, with column file_path

required

Kwargs: return_proba (bool): If the function should return the probabilities instead of the classes -- classifier only batch_size (int): Batch size to be used -- classifier only Raises: NotImplementedError: If the model is not classifier nor object_detector Returns: (np.ndarray | list): Array, shape = [n_samples, n_classes] or List of n_samples elements

Source code in template_vision/models_training/model_keras.py
@utils.trained_needed
def predict(self, df_test: pd.DataFrame, return_proba: bool = False, batch_size: Union[int, None] = None) -> Union[np.ndarray, list]:
    '''Predictions on test set

    Args:
        df_test (pd.DataFrame): DataFrame to be predicted, with column file_path
    Kwargs:
        return_proba (bool): If the function should return the probabilities instead of the classes -- classifier only
        batch_size (int): Batch size to be used -- classifier only
    Raises:
        NotImplementedError: If the model is not `classifier` nor `object_detector`
    Returns:
        (np.ndarray | list): Array, shape = [n_samples, n_classes] or List of n_samples elements
    '''
    if self.model_type == 'classifier':
        return self._predict_classifier(df_test, return_proba=return_proba, batch_size=batch_size)
    elif self.model_type == 'object_detector':
        return self._predict_object_detector(df_test)
    else:
        raise NotImplementedError("Only 'classifier' and 'object_detector' model type are supported")

predict_proba(df_test, batch_size=None)

Predicts probabilities on the test dataset

Parameters:

Name Type Description Default
df_test DataFrame

DataFrame to be predicted, with column file_path

required

Kwargs: batch_size (int): Batch size to be used Raises: ValueError: If the model is not a classifier Returns: (np.ndarray): Array, shape = [n_samples, n_classes]

Source code in template_vision/models_training/model_keras.py
@utils.trained_needed
def predict_proba(self, df_test, batch_size: int = None) -> np.ndarray:
    '''Predicts probabilities on the test dataset

    Args:
        df_test (pd.DataFrame): DataFrame to be predicted, with column file_path
    Kwargs:
        batch_size (int): Batch size to be used
    Raises:
        ValueError: If the model is not a classifier
    Returns:
        (np.ndarray): Array, shape = [n_samples, n_classes]
    '''
    if self.model_type != 'classifier':
        raise ValueError(f"`predict_proba` function does not support model type {self.model_type}")

    # We reuse the predict function
    return self.predict(df_test, return_proba=True, batch_size=batch_size)

reload_from_standalone(**kwargs)

Reloads a model from its configuration and "standalones" files - /! Needs to be overridden /! -

Source code in template_vision/models_training/model_keras.py
def reload_from_standalone(self, **kwargs) -> None:
    '''Reloads a model from its configuration and "standalones" files
    - /!\\ Needs to be overridden /!\\ -
    '''
    raise NotImplementedError("'reload' needs to be overridden")

reload_model(hdf5_path)

Loads a Keras model from a HDF5 file

Parameters:

Name Type Description Default
hdf5_path str

Path to the hdf5 file

required

Returns: ?: Keras model

Source code in template_vision/models_training/model_keras.py
def reload_model(self, hdf5_path: str) -> Any:
    '''Loads a Keras model from a HDF5 file

    Args:
        hdf5_path (str): Path to the hdf5 file
    Returns:
        ?: Keras model
    '''
    # Fix tensorflow GPU if not already done (useful if we reload a model)
    try:
        gpu_devices = tf.config.experimental.list_physical_devices('GPU')
        for device in gpu_devices:
            tf.config.experimental.set_memory_growth(device, True)
    except Exception:
        pass

    # We check if we already have the custom objects
    if hasattr(self, 'custom_objects') and self.custom_objects is not None:
        custom_objects = self.custom_objects
    else:
        self.logger.warning("Can't find the attribute 'custom_objects' in the model to be reloaded")
        self.logger.warning("Backup on the default custom_objects of utils_deep_keras")
        custom_objects = utils_deep_keras.custom_objects

    # Loading of the model
    keras_model = load_model(hdf5_path, custom_objects=custom_objects)

    # Set trained to true if not already true
    if not self.trained:
        self.trained = True
        self.nb_fit = 1

    # Return
    return keras_model

save(json_data=None)

Saves the model

Kwargs

json_data (dict): Additional configurations to be saved

Source code in template_vision/models_training/model_keras.py
@no_type_check  # We do not check the type, because it is complicated with managing custom_objects_str
def save(self, json_data: Union[dict, None] = None) -> None:
    '''Saves the model

    Kwargs:
        json_data (dict): Additional configurations to be saved
    '''
    # Save configuration JSON
    if json_data is None:
        json_data = {}

    json_data['librairie'] = 'keras'
    json_data['batch_size'] = self.batch_size
    json_data['epochs'] = self.epochs
    json_data['validation_split'] = self.validation_split
    json_data['patience'] = self.patience
    json_data['width'] = self.width
    json_data['height'] = self.height
    json_data['depth'] = self.depth
    json_data['color_mode'] = self.color_mode
    json_data['in_memory'] = self.in_memory
    json_data['data_augmentation_params'] = self.data_augmentation_params
    json_data['nb_train_generator_images_to_save'] = self.nb_train_generator_images_to_save
    json_data['keras_params'] = self.keras_params
    if self.model is not None:
        json_data['keras_model'] = json.loads(self.model.to_json())
    else:
        json_data['keras_model'] = None

    # Add _get_model code if not in json_data
    if '_get_model' not in json_data.keys():
        json_data['_get_model'] = pickle.source.getsourcelines(self._get_model)[0]
    # Add _get_preprocess_input code if not in json_data
    if '_get_preprocess_input' not in json_data.keys():
        json_data['_get_preprocess_input'] = pickle.source.getsourcelines(self._get_preprocess_input)[0]
    # Save preprocess_input to a .pkl file if level_save > LOW
    pkl_path = os.path.join(self.model_dir, "preprocess_input.pkl")
    if self.level_save in ['MEDIUM', 'HIGH']:
        with open(pkl_path, 'wb') as f:
            pickle.dump(self.preprocess_input, f)
    # Add _get_learning_rate_scheduler code if not in json_data
    if '_get_learning_rate_scheduler' not in json_data.keys():
        json_data['_get_learning_rate_scheduler'] = pickle.source.getsourcelines(self._get_learning_rate_scheduler)[0]
    # Add custom_objects code if not in json_data
    if 'custom_objects' not in json_data.keys():
        custom_objects_str = self.custom_objects.copy()
        for key in custom_objects_str.keys():
            if callable(custom_objects_str[key]):
                # Nominal case
                if not type(custom_objects_str[key]) == functools.partial:
                    custom_objects_str[key] = pickle.source.getsourcelines(custom_objects_str[key])[0]
                # Manage partials
                else:
                    custom_objects_str[key] = {
                        'type': 'partial',
                        'args': custom_objects_str[key].args,
                        'function': pickle.source.getsourcelines(custom_objects_str[key].func)[0],
                    }
        json_data['custom_objects'] = custom_objects_str

    # Save strategy :
    # - best.hdf5 already saved in fit()
    # - can't pickle keras model, so we drop it, save, and reload it
    keras_model = self.model
    self.model = None
    super().save(json_data=json_data)
    self.model = keras_model