train_utils.py 21.7 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
# @author lucasmiranda42
# encoding: utf-8
# module deepof

"""

Simple utility functions used in deepof example scripts. These are not part of the main package

"""

lucas_miranda's avatar
lucas_miranda committed
11
12
import json
import os
13
from datetime import date, datetime
lucas_miranda's avatar
lucas_miranda committed
14
15
16
17
from typing import Tuple, Union, Any, List

import numpy as np
import tensorflow as tf
18
from kerastuner import BayesianOptimization, Hyperband
19
from kerastuner_tensorboard_logger import TensorBoardLogger
lucas_miranda's avatar
lucas_miranda committed
20
from sklearn.metrics import roc_auc_score
21
from tensorboard.plugins.hparams import api as hp
lucas_miranda's avatar
lucas_miranda committed
22

23
24
25
import deepof.hypermodels
import deepof.model_utils

26
27
28
29
30
# Ignore warning with no downstream effect
tf.get_logger().setLevel("ERROR")
tf.autograph.set_verbosity(0)


31
class CustomStopper(tf.keras.callbacks.EarlyStopping):
32
    """ Custom early stopping callback. Prevents the model from stopping before warmup is over """
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49

    def __init__(self, start_epoch, *args, **kwargs):
        super(CustomStopper, self).__init__(*args, **kwargs)
        self.start_epoch = start_epoch

    def get_config(self):  # pragma: no cover
        """Updates callback metadata"""

        config = super().get_config().copy()
        config.update({"start_epoch": self.start_epoch})
        return config

    def on_epoch_end(self, epoch, logs=None):
        if epoch > self.start_epoch:
            super().on_epoch_end(epoch, logs)


50
51
52
53
54
def load_treatments(train_path):
    """Loads a dictionary containing the treatments per individual,
    to be loaded as metadata in the coordinates class"""
    try:
        with open(
55
56
57
58
59
            os.path.join(
                train_path,
                [i for i in os.listdir(train_path) if i.endswith(".json")][0],
            ),
            "r",
60
        ) as handle:
61
            treatment_dict = json.load(handle)
62
63
64
65
66
67
68
    except IndexError:
        treatment_dict = None

    return treatment_dict


def get_callbacks(
69
70
71
    X_train: np.array,
    batch_size: int,
    variational: bool,
72
73
74
    phenotype_prediction: float,
    next_sequence_prediction: float,
    rule_based_prediction: float,
75
76
    loss: str,
    X_val: np.array = None,
77
    input_type: str = False,
78
79
80
81
82
83
84
    cp: bool = False,
    reg_cat_clusters: bool = False,
    reg_cluster_variance: bool = False,
    entropy_samples: int = 15000,
    entropy_knn: int = 100,
    logparam: dict = None,
    outpath: str = ".",
85
    run: int = False,
86
) -> List[Union[Any]]:
87
    """Generates callbacks for model training, including:
88
89
    - run_ID: run name, with coarse parameter details;
    - tensorboard_callback: for real-time visualization;
90
91
92
93
    - cp_callback: for checkpoint saving;
    - onecycle: for learning rate scheduling;
    - entropy: neighborhood entropy in the latent space;
    """
94

95
96
97
98
99
100
101
102
    latreg = "none"
    if reg_cat_clusters and not reg_cluster_variance:
        latreg = "categorical"
    elif reg_cluster_variance and not reg_cat_clusters:
        latreg = "variance"
    elif reg_cat_clusters and reg_cluster_variance:
        latreg = "categorical+variance"

103
    run_ID = "{}{}{}{}{}{}{}{}{}{}{}{}{}".format(
104
        ("GMVAE" if variational else "AE"),
105
        ("_input_type={}".format(input_type) if input_type else "coords"),
106
        ("_window_size={}".format(X_train.shape[1])),
107
108
109
        ("_NextSeqPred={}".format(next_sequence_prediction) if variational else ""),
        ("_PhenoPred={}".format(phenotype_prediction) if variational else ""),
        ("_RuleBasedPred={}".format(rule_based_prediction) if variational else ""),
110
        ("_loss={}".format(loss) if variational else ""),
111
112
        ("_encoding={}".format(logparam["encoding"]) if logparam is not None else ""),
        ("_k={}".format(logparam["k"]) if logparam is not None else ""),
113
        ("_latreg={}".format(latreg)),
114
115
        ("_entknn={}".format(entropy_knn)),
        ("_run={}".format(run) if run else ""),
116
        ("_{}".format(datetime.now().strftime("%Y%m%d-%H%M%S")) if not run else ""),
117
118
    )

119
    log_dir = os.path.abspath(os.path.join(outpath, "fit", run_ID))
120
    tensorboard_callback = tf.keras.callbacks.TensorBoard(
121
122
123
        log_dir=log_dir,
        histogram_freq=1,
        profile_batch=2,
124
125
    )

126
    entropy = deepof.model_utils.neighbor_latent_entropy(
127
        encoding_dim=logparam["encoding"],
128
        k=entropy_knn,
129
        samples=entropy_samples,
lucas_miranda's avatar
lucas_miranda committed
130
        validation_data=X_val,
131
        log_dir=os.path.join(outpath, "metrics", run_ID),
132
        variational=variational,
lucas_miranda's avatar
lucas_miranda committed
133
134
    )

135
    onecycle = deepof.model_utils.one_cycle_scheduler(
136
137
        X_train.shape[0] // batch_size * 250,
        max_rate=0.005,
138
        log_dir=os.path.join(outpath, "metrics", run_ID),
139
140
    )

141
    callbacks = [run_ID, tensorboard_callback, entropy, onecycle]
142
143
144

    if cp:
        cp_callback = tf.keras.callbacks.ModelCheckpoint(
145
            os.path.join(outpath, "checkpoints", run_ID + "/cp-{epoch:04d}.ckpt"),
146
147
148
149
150
151
152
153
            verbose=1,
            save_best_only=False,
            save_weights_only=True,
            save_freq="epoch",
        )
        callbacks.append(cp_callback)

    return callbacks
154
155


lucas_miranda's avatar
lucas_miranda committed
156
def log_hyperparameters(phenotype_class: float, rec: str):
lucas_miranda's avatar
lucas_miranda committed
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
    """Blueprint for hyperparameter and metric logging in tensorboard during hyperparameter tuning"""

    logparams = [
        hp.HParam(
            "encoding",
            hp.Discrete([2, 4, 6, 8, 12, 16]),
            display_name="encoding",
            description="encoding size dimensionality",
        ),
        hp.HParam(
            "k",
            hp.IntInterval(min_value=1, max_value=25),
            display_name="k",
            description="cluster_number",
        ),
        hp.HParam(
            "loss",
            hp.Discrete(["ELBO", "MMD", "ELBO+MMD"]),
            display_name="loss function",
            description="loss function",
        ),
    ]

    metrics = [
        hp.Metric("val_{}mae".format(rec), display_name="val_{}mae".format(rec)),
        hp.Metric("val_{}mse".format(rec), display_name="val_{}mse".format(rec)),
    ]
    if phenotype_class:
        logparams.append(
            hp.HParam(
                "pheno_weight",
                hp.RealInterval(min_value=0.0, max_value=1000.0),
                display_name="pheno weight",
                description="weight applied to phenotypic classifier from the latent space",
            )
        )
        metrics += [
            hp.Metric(
                "phenotype_prediction_accuracy",
                display_name="phenotype_prediction_accuracy",
            ),
            hp.Metric(
                "phenotype_prediction_auc",
                display_name="phenotype_prediction_auc",
            ),
        ]

    return logparams, metrics


# noinspection PyUnboundLocalVariable
lucas_miranda's avatar
lucas_miranda committed
208
def tensorboard_metric_logging(
209
210
211
212
213
    run_dir: str,
    hpms: Any,
    ae: Any,
    X_val: np.ndarray,
    y_val: np.ndarray,
214
215
216
    next_sequence_prediction: float,
    phenotype_prediction: float,
    rule_based_prediction: float,
217
    rec: str,
lucas_miranda's avatar
lucas_miranda committed
218
):
lucas_miranda's avatar
lucas_miranda committed
219
220
    """Autoencoder metric logging in tensorboard"""

221
222
    outputs = ae.predict(X_val)
    idx_generator = (idx for idx in range(len(outputs)))
lucas_miranda's avatar
lucas_miranda committed
223
224
225

    with tf.summary.create_file_writer(run_dir).as_default():
        hp.hparams(hpms)  # record the values used in this trial
226
227
        idx = next(idx_generator)

lucas_miranda's avatar
lucas_miranda committed
228
        val_mae = tf.reduce_mean(
229
            tf.keras.metrics.mean_absolute_error(y_val[idx], outputs[idx])
lucas_miranda's avatar
lucas_miranda committed
230
231
        )
        val_mse = tf.reduce_mean(
232
            tf.keras.metrics.mean_squared_error(y_val[idx], outputs[idx])
lucas_miranda's avatar
lucas_miranda committed
233
234
235
236
        )
        tf.summary.scalar("val_{}mae".format(rec), val_mae, step=1)
        tf.summary.scalar("val_{}mse".format(rec), val_mse, step=1)

237
        if next_sequence_prediction:
238
            idx = next(idx_generator)
lucas_miranda's avatar
lucas_miranda committed
239
            pred_mae = tf.reduce_mean(
240
                tf.keras.metrics.mean_absolute_error(y_val[idx], outputs[idx])
lucas_miranda's avatar
lucas_miranda committed
241
242
            )
            pred_mse = tf.reduce_mean(
243
                tf.keras.metrics.mean_squared_error(y_val[idx], outputs[idx])
244
245
246
247
248
249
            )
            tf.summary.scalar(
                "val_next_sequence_prediction_mae".format(rec), pred_mae, step=1
            )
            tf.summary.scalar(
                "val_next_sequence_prediction_mse".format(rec), pred_mse, step=1
lucas_miranda's avatar
lucas_miranda committed
250
251
            )

252
        if phenotype_prediction:
253
            idx = next(idx_generator)
254
255
256
            pheno_acc = tf.keras.metrics.binary_accuracy(
                y_val[idx], tf.squeeze(outputs[idx])
            )
257
            pheno_auc = tf.keras.metrics.AUC()
258
            pheno_auc.update_state(y_val[idx], outputs[idx])
259
            pheno_auc = pheno_auc.result().numpy()
lucas_miranda's avatar
lucas_miranda committed
260
261
262
263

            tf.summary.scalar("phenotype_prediction_accuracy", pheno_acc, step=1)
            tf.summary.scalar("phenotype_prediction_auc", pheno_auc, step=1)

264
        if rule_based_prediction:
265
            idx = next(idx_generator)
266
            rules_mae = tf.reduce_mean(
267
                tf.keras.metrics.mean_absolute_error(y_val[idx], outputs[idx])
268
269
            )
            rules_mse = tf.reduce_mean(
270
                tf.keras.metrics.mean_squared_error(y_val[idx], outputs[idx])
271
272
273
274
            )
            tf.summary.scalar("val_prediction_mae".format(rec), rules_mae, step=1)
            tf.summary.scalar("val_prediction_mse".format(rec), rules_mse, step=1)

lucas_miranda's avatar
lucas_miranda committed
275

276
def autoencoder_fitting(
277
278
279
280
281
    preprocessed_object: Tuple[np.ndarray, np.ndarray, np.ndarray, np.ndarray],
    batch_size: int,
    encoding_size: int,
    epochs: int,
    hparams: dict,
282
    kl_annealing_mode: str,
283
284
285
286
    kl_warmup: int,
    log_history: bool,
    log_hparams: bool,
    loss: str,
287
    mmd_annealing_mode: str,
288
289
290
291
    mmd_warmup: int,
    montecarlo_kl: int,
    n_components: int,
    output_path: str,
292
293
294
    next_sequence_prediction: float,
    phenotype_prediction: float,
    rule_based_prediction: float,
295
296
297
298
299
300
301
302
    pretrained: str,
    save_checkpoints: bool,
    save_weights: bool,
    variational: bool,
    reg_cat_clusters: bool,
    reg_cluster_variance: bool,
    entropy_samples: int,
    entropy_knn: int,
303
    input_type: str,
304
    run: int = 0,
305
):
306
307
    """Implementation function for deepof.data.coordinates.deep_unsupervised_embedding"""

308
    # Load data
309
310
311
312
313
    X_train, y_train, X_val, y_val = preprocessed_object

    # To avoid stability issues
    tf.keras.backend.clear_session()

314
    # Defines what to log on tensorboard (useful for trying out different models)
315
316
    logparam = {
        "encoding": encoding_size,
317
        "k": n_components,
318
319
        "loss": loss,
    }
320
321
    if phenotype_prediction:
        logparam["pheno_weight"] = phenotype_prediction
322

323
    # Load callbacks
324
    run_ID, *cbacks = get_callbacks(
325
326
327
        X_train=X_train,
        batch_size=batch_size,
        variational=variational,
328
        phenotype_prediction=phenotype_prediction,
329
        next_sequence_prediction=next_sequence_prediction,
330
        rule_based_prediction=rule_based_prediction,
331
        loss=loss,
332
333
334
        input_type=input_type,
        X_val=(X_val if X_val.shape != (0,) else None),
        cp=save_checkpoints,
335
        reg_cat_clusters=reg_cat_clusters,
336
        reg_cluster_variance=reg_cluster_variance,
337
338
        entropy_samples=entropy_samples,
        entropy_knn=entropy_knn,
339
340
        logparam=logparam,
        outpath=output_path,
341
        run=run,
342
    )
343
344
    if not log_history:
        cbacks = cbacks[1:]
345

346
    # Logs hyperparameters to tensorboard
347
    rec = "reconstruction_" if phenotype_prediction else ""
348
    if log_hparams:
349
        logparams, metrics = log_hyperparameters(phenotype_prediction, rec)
350
351

        with tf.summary.create_file_writer(
352
            os.path.join(output_path, "hparams", run_ID)
353
354
355
356
357
        ).as_default():
            hp.hparams_config(
                hparams=logparams,
                metrics=metrics,
            )
358

359
360
361
362
363
364
365
366
    # Gets the number of rule-based features
    try:
        rule_based_features = (
            y_train.shape[1] if not phenotype_prediction else y_train.shape[1] - 1
        )
    except IndexError:
        rule_based_features = 0

367
    # Build models
368
369
370
371
372
373
374
    if not variational:
        encoder, decoder, ae = deepof.models.SEQ_2_SEQ_AE(
            ({} if hparams is None else hparams)
        ).build(X_train.shape)
        return_list = (encoder, decoder, ae)

    else:
375
376
377
378
379
380
381
382
        (
            encoder,
            generator,
            grouper,
            ae,
            prior,
            posterior,
        ) = deepof.models.SEQ_2_SEQ_GMVAE(
383
384
385
386
            architecture_hparams=({} if hparams is None else hparams),
            batch_size=batch_size,
            compile_model=True,
            encoding=encoding_size,
387
            kl_annealing_mode=kl_annealing_mode,
388
389
            kl_warmup_epochs=kl_warmup,
            loss=loss,
390
            mmd_annealing_mode=mmd_annealing_mode,
391
392
393
394
395
            mmd_warmup_epochs=mmd_warmup,
            montecarlo_kl=montecarlo_kl,
            neuron_control=False,
            number_of_components=n_components,
            overlap_loss=False,
396
397
398
            next_sequence_prediction=next_sequence_prediction,
            phenotype_prediction=phenotype_prediction,
            rule_based_prediction=rule_based_prediction,
399
            rule_based_features=rule_based_features,
400
401
            reg_cat_clusters=reg_cat_clusters,
            reg_cluster_variance=reg_cluster_variance,
402
403
404
        ).build(
            X_train.shape
        )
405
406
407
        return_list = (encoder, generator, grouper, ae)

    if pretrained:
408
        # If pretrained models are specified, load weights and return
409
410
411
412
413
414
415
416
417
        ae.load_weights(pretrained)
        return return_list

    else:
        if not variational:

            ae.fit(
                x=X_train,
                y=X_train,
418
                epochs=epochs,
419
420
421
                batch_size=batch_size,
                verbose=1,
                validation_data=(X_val, X_val),
422
                callbacks=cbacks
423
424
425
426
427
428
429
430
                + [
                    CustomStopper(
                        monitor="val_loss",
                        patience=5,
                        restore_best_weights=True,
                        start_epoch=max(kl_warmup, mmd_warmup),
                    ),
                ],
431
432
            )

433
434
435
            if not os.path.exists(os.path.join(output_path, "trained_weights")):
                os.makedirs(os.path.join(output_path, "trained_weights"))

436
            if save_weights:
437
438
439
440
441
442
443
                ae.save_weights(
                    os.path.join(
                        "{}".format(output_path),
                        "trained_weights",
                        "{}_final_weights.h5".format(run_ID),
                    )
                )
444

445
446
        else:

447
            callbacks_ = cbacks + [
448
449
450
451
                CustomStopper(
                    monitor="val_loss",
                    patience=5,
                    restore_best_weights=True,
452
                    start_epoch=max(kl_warmup, mmd_warmup),
453
454
455
                ),
            ]

456
457
            Xs, ys = X_train, [X_train]
            Xvals, yvals = X_val, [X_val]
458

459
            if next_sequence_prediction > 0.0:
460
461
462
                Xs, ys = X_train[:-1], [X_train[:-1], X_train[1:]]
                Xvals, yvals = X_val[:-1], [X_val[:-1], X_val[1:]]

463
            if phenotype_prediction > 0.0:
464
                ys += [y_train[-Xs.shape[0] :, 0]]
465
                yvals += [y_val[-Xvals.shape[0] :, 0]]
466
467
468
469
470
471

                # Remove the used column (phenotype) from both y arrays
                y_train = y_train[:, 1:]
                y_val = y_val[:, 1:]

            if rule_based_prediction > 0.0:
472
                ys += [y_train[-Xs.shape[0] :]]
473
                yvals += [y_val[-Xvals.shape[0] :]]
474

475
            ae.fit(
476
477
                x=Xs,
                y=ys,
478
                epochs=epochs,
479
480
481
482
483
484
485
486
487
                batch_size=batch_size,
                verbose=1,
                validation_data=(
                    Xvals,
                    yvals,
                ),
                callbacks=callbacks_,
            )

488
            if not os.path.exists(os.path.join(output_path, "trained_weights")):
489
                os.makedirs(os.path.join(output_path, "trained_weights"))
490

491
            if save_weights:
492
493
                ae.save_weights(
                    os.path.join(
494
495
496
                        "{}".format(output_path),
                        "trained_weights",
                        "{}_final_weights.h5".format(run_ID),
497
498
                    )
                )
499

500
501
502
            if log_hparams:
                # Logparams to tensorboard
                tensorboard_metric_logging(
503
504
505
506
507
508
509
510
511
                    run_dir=os.path.join(output_path, "hparams", run_ID),
                    hpms=logparam,
                    ae=ae,
                    X_val=Xvals,
                    y_val=yvals,
                    next_sequence_prediction=next_sequence_prediction,
                    phenotype_prediction=phenotype_prediction,
                    rule_based_prediction=rule_based_prediction,
                    rec=rec,
512
                )
513

514
515
516
    return return_list


517
def tune_search(
518
519
520
521
522
523
524
525
526
527
    data: List[np.array],
    encoding_size: int,
    hypertun_trials: int,
    hpt_type: str,
    hypermodel: str,
    k: int,
    kl_warmup_epochs: int,
    loss: str,
    mmd_warmup_epochs: int,
    overlap_loss: float,
528
529
530
    next_sequence_prediction: float,
    phenotype_prediction: float,
    rule_based_prediction: float,
531
532
533
534
535
    project_name: str,
    callbacks: List,
    n_epochs: int = 30,
    n_replicas: int = 1,
    outpath: str = ".",
536
) -> Union[bool, Tuple[Any, Any]]:
537
538
    """Define the search space using keras-tuner and bayesian optimization

539
540
541
542
543
544
545
546
547
548
549
    Parameters:
        - train (np.array): dataset to train the model on
        - test (np.array): dataset to validate the model on
        - hypertun_trials (int): number of Bayesian optimization iterations to run
        - hpt_type (str): specify one of Bayesian Optimization (bayopt) and Hyperband (hyperband)
        - hypermodel (str): hypermodel to load. Must be one of S2SAE (plain autoencoder)
        or S2SGMVAE (Gaussian Mixture Variational autoencoder).
        - k (int) number of components of the Gaussian Mixture
        - loss (str): one of [ELBO, MMD, ELBO+MMD]
        - overlap_loss (float): assigns as weight to an extra loss term which
        penalizes overlap between GM components
550
        - phenotype_class (float): adds an extra regularizing neural network to the model,
551
552
553
554
555
556
557
558
559
560
561
562
        which tries to predict the phenotype of the animal from which the sequence comes
        - predictor (float): adds an extra regularizing neural network to the model,
        which tries to predict the next frame from the current one
        - project_name (str): ID of the current run
        - callbacks (list): list of callbacks for the training loop
        - n_epochs (int): optional. Number of epochs to train each run for
        - n_replicas (int): optional. Number of replicas per parameter set. Higher values
         will yield more robust results, but will affect performance severely

    Returns:
        - best_hparams (dict): dictionary with the best retrieved hyperparameters
        - best_run (tf.keras.Model): trained instance of the best model found
563
564
565

    """

566
567
    X_train, y_train, X_val, y_val = data

568
569
570
571
    assert hpt_type in ["bayopt", "hyperband"], (
        "Invalid hyperparameter tuning framework. " "Select one of bayopt and hyperband"
    )

lucas_miranda's avatar
lucas_miranda committed
572
    if hypermodel == "S2SAE":  # pragma: no cover
573
        assert (
574
            next_sequence_prediction == 0.0 and phenotype_prediction == 0.0
575
        ), "Prediction branches are only available for variational models. See documentation for more details"
576
        batch_size = 1
577
        hypermodel = deepof.hypermodels.SEQ_2_SEQ_AE(input_shape=X_train.shape)
578
579

    elif hypermodel == "S2SGMVAE":
580
        batch_size = 64
581
        hypermodel = deepof.hypermodels.SEQ_2_SEQ_GMVAE(
582
            input_shape=X_train.shape,
583
            encoding=encoding_size,
584
            kl_warmup_epochs=kl_warmup_epochs,
585
            loss=loss,
586
            mmd_warmup_epochs=mmd_warmup_epochs,
587
            number_of_components=k,
588
            overlap_loss=overlap_loss,
589
590
591
            next_sequence_prediction=next_sequence_prediction,
            phenotype_prediction=phenotype_prediction,
            rule_based_prediction=rule_based_prediction,
592
593
594
            rule_based_features=(
                y_train.shape[1] if not phenotype_prediction else y_train.shape[1] - 1
            ),
595
        )
lucas_miranda's avatar
lucas_miranda committed
596

597
598
599
    else:
        return False

600
601
602
    hpt_params = {
        "hypermodel": hypermodel,
        "executions_per_trial": n_replicas,
603
604
605
        "logger": TensorBoardLogger(
            metrics=["val_mae"], logdir=os.path.join(outpath, "logged_hparams")
        ),
606
607
608
609
610
611
612
        "objective": "val_mae",
        "project_name": project_name,
        "tune_new_entries": True,
    }

    if hpt_type == "hyperband":
        tuner = Hyperband(
613
614
615
            directory=os.path.join(
                outpath, "HyperBandx_{}_{}".format(loss, str(date.today()))
            ),
616
            max_epochs=30,
617
            hyperband_iterations=hypertun_trials,
618
            factor=3,
619
620
621
622
            **hpt_params
        )
    else:
        tuner = BayesianOptimization(
623
624
625
            directory=os.path.join(
                outpath, "BayOpt_{}_{}".format(loss, str(date.today()))
            ),
626
627
628
            max_trials=hypertun_trials,
            **hpt_params
        )
629
630
631

    print(tuner.search_space_summary())

632
633
634
    Xs, ys = [X_train], [X_train]
    Xvals, yvals = [X_val], [X_val]

635
    if next_sequence_prediction > 0.0:
636
637
638
        Xs, ys = X_train[:-1], [X_train[:-1], X_train[1:]]
        Xvals, yvals = X_val[:-1], [X_val[:-1], X_val[1:]]

639
    if phenotype_prediction > 0.0:
640
641
        ys += [y_train[-Xs.shape[0] :, 0]]
        yvals += [y_val[-Xvals.shape[0] :, 0]]
642
643
644
645
646
647

        # Remove the used column (phenotype) from both y arrays
        y_train = y_train[:, 1:]
        y_val = y_val[:, 1:]

    if rule_based_prediction > 0.0:
648
649
        ys += [y_train[-Xs.shape[0] :]]
        yvals += [y_val[-Xvals.shape[0] :]]
650

651
    tuner.search(
652
653
        Xs,
        ys,
654
        epochs=n_epochs,
655
        validation_data=(Xvals, yvals),
656
        verbose=1,
657
        batch_size=batch_size,
lucas_miranda's avatar
lucas_miranda committed
658
        callbacks=callbacks,
659
660
661
662
663
    )

    best_hparams = tuner.get_best_hyperparameters(num_trials=1)[0]
    best_run = tuner.hypermodel.build(best_hparams)

lucas_miranda's avatar
lucas_miranda committed
664
665
    print(tuner.results_summary())

666
    return best_hparams, best_run