train_utils.py 21.5 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
282
283
284
285
286
287
288
289
    preprocessed_object: Tuple[np.ndarray, np.ndarray, np.ndarray, np.ndarray],
    batch_size: int,
    encoding_size: int,
    epochs: int,
    hparams: dict,
    kl_warmup: int,
    log_history: bool,
    log_hparams: bool,
    loss: str,
    mmd_warmup: int,
    montecarlo_kl: int,
    n_components: int,
    output_path: str,
290
291
292
    next_sequence_prediction: float,
    phenotype_prediction: float,
    rule_based_prediction: float,
293
294
295
296
297
298
299
300
    pretrained: str,
    save_checkpoints: bool,
    save_weights: bool,
    variational: bool,
    reg_cat_clusters: bool,
    reg_cluster_variance: bool,
    entropy_samples: int,
    entropy_knn: int,
301
    input_type: str,
302
    run: int = 0,
303
):
304
305
    """Implementation function for deepof.data.coordinates.deep_unsupervised_embedding"""

306
    # Load data
307
308
309
310
311
    X_train, y_train, X_val, y_val = preprocessed_object

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

312
    # Defines what to log on tensorboard (useful for trying out different models)
313
314
    logparam = {
        "encoding": encoding_size,
315
        "k": n_components,
316
317
        "loss": loss,
    }
318
319
    if phenotype_prediction:
        logparam["pheno_weight"] = phenotype_prediction
320

321
    # Load callbacks
322
    run_ID, *cbacks = get_callbacks(
323
324
325
        X_train=X_train,
        batch_size=batch_size,
        variational=variational,
326
        phenotype_prediction=phenotype_prediction,
327
        next_sequence_prediction=next_sequence_prediction,
328
        rule_based_prediction=rule_based_prediction,
329
        loss=loss,
330
331
332
        input_type=input_type,
        X_val=(X_val if X_val.shape != (0,) else None),
        cp=save_checkpoints,
333
        reg_cat_clusters=reg_cat_clusters,
334
        reg_cluster_variance=reg_cluster_variance,
335
336
        entropy_samples=entropy_samples,
        entropy_knn=entropy_knn,
337
338
        logparam=logparam,
        outpath=output_path,
339
        run=run,
340
    )
341
342
    if not log_history:
        cbacks = cbacks[1:]
343

344
    # Logs hyperparameters to tensorboard
345
    rec = "reconstruction_" if phenotype_prediction else ""
346
    if log_hparams:
347
        logparams, metrics = log_hyperparameters(phenotype_prediction, rec)
348
349

        with tf.summary.create_file_writer(
350
            os.path.join(output_path, "hparams", run_ID)
351
352
353
354
355
        ).as_default():
            hp.hparams_config(
                hparams=logparams,
                metrics=metrics,
            )
356

357
358
359
360
361
362
363
364
    # 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

365
    # Build models
366
367
368
369
370
371
372
    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:
373
374
375
376
377
378
379
380
        (
            encoder,
            generator,
            grouper,
            ae,
            prior,
            posterior,
        ) = deepof.models.SEQ_2_SEQ_GMVAE(
381
382
383
384
385
386
387
388
389
390
391
            architecture_hparams=({} if hparams is None else hparams),
            batch_size=batch_size,
            compile_model=True,
            encoding=encoding_size,
            kl_warmup_epochs=kl_warmup,
            loss=loss,
            mmd_warmup_epochs=mmd_warmup,
            montecarlo_kl=montecarlo_kl,
            neuron_control=False,
            number_of_components=n_components,
            overlap_loss=False,
392
393
394
            next_sequence_prediction=next_sequence_prediction,
            phenotype_prediction=phenotype_prediction,
            rule_based_prediction=rule_based_prediction,
395
            rule_based_features=rule_based_features,
396
397
            reg_cat_clusters=reg_cat_clusters,
            reg_cluster_variance=reg_cluster_variance,
398
399
400
        ).build(
            X_train.shape
        )
401
402
403
        return_list = (encoder, generator, grouper, ae)

    if pretrained:
404
        # If pretrained models are specified, load weights and return
405
406
407
408
409
410
411
412
413
        ae.load_weights(pretrained)
        return return_list

    else:
        if not variational:

            ae.fit(
                x=X_train,
                y=X_train,
414
                epochs=epochs,
415
416
417
                batch_size=batch_size,
                verbose=1,
                validation_data=(X_val, X_val),
418
                callbacks=cbacks
419
420
421
422
423
424
425
426
                + [
                    CustomStopper(
                        monitor="val_loss",
                        patience=5,
                        restore_best_weights=True,
                        start_epoch=max(kl_warmup, mmd_warmup),
                    ),
                ],
427
428
            )

429
430
431
            if not os.path.exists(os.path.join(output_path, "trained_weights")):
                os.makedirs(os.path.join(output_path, "trained_weights"))

432
            if save_weights:
433
434
435
436
437
438
439
                ae.save_weights(
                    os.path.join(
                        "{}".format(output_path),
                        "trained_weights",
                        "{}_final_weights.h5".format(run_ID),
                    )
                )
440

441
442
        else:

443
            callbacks_ = cbacks + [
444
445
446
447
                CustomStopper(
                    monitor="val_loss",
                    patience=5,
                    restore_best_weights=True,
448
                    start_epoch=max(kl_warmup, mmd_warmup),
449
450
451
                ),
            ]

452
453
            Xs, ys = X_train, [X_train]
            Xvals, yvals = X_val, [X_val]
454

455
            if next_sequence_prediction > 0.0:
456
457
458
                Xs, ys = X_train[:-1], [X_train[:-1], X_train[1:]]
                Xvals, yvals = X_val[:-1], [X_val[:-1], X_val[1:]]

459
            if phenotype_prediction > 0.0:
460
                ys += [y_train[-Xs.shape[0] :, 0]]
461
                yvals += [y_val[-Xvals.shape[0] :, 0]]
462
463
464
465
466
467

                # 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:
468
                ys += [y_train[-Xs.shape[0] :]]
469
                yvals += [y_val[-Xvals.shape[0] :]]
470

471
            ae.fit(
472
473
                x=Xs,
                y=ys,
474
                epochs=epochs,
475
476
477
478
479
480
481
482
483
                batch_size=batch_size,
                verbose=1,
                validation_data=(
                    Xvals,
                    yvals,
                ),
                callbacks=callbacks_,
            )

484
            if not os.path.exists(os.path.join(output_path, "trained_weights")):
485
                os.makedirs(os.path.join(output_path, "trained_weights"))
486

487
            if save_weights:
488
489
                ae.save_weights(
                    os.path.join(
490
491
492
                        "{}".format(output_path),
                        "trained_weights",
                        "{}_final_weights.h5".format(run_ID),
493
494
                    )
                )
495

496
497
498
            if log_hparams:
                # Logparams to tensorboard
                tensorboard_metric_logging(
499
500
501
502
503
504
505
506
507
                    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,
508
                )
509

510
511
512
    return return_list


513
def tune_search(
514
515
516
517
518
519
520
521
522
523
    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,
524
525
526
    next_sequence_prediction: float,
    phenotype_prediction: float,
    rule_based_prediction: float,
527
528
529
530
531
    project_name: str,
    callbacks: List,
    n_epochs: int = 30,
    n_replicas: int = 1,
    outpath: str = ".",
532
) -> Union[bool, Tuple[Any, Any]]:
533
534
    """Define the search space using keras-tuner and bayesian optimization

535
536
537
538
539
540
541
542
543
544
545
    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
546
        - phenotype_class (float): adds an extra regularizing neural network to the model,
547
548
549
550
551
552
553
554
555
556
557
558
        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
559
560
561

    """

562
563
    X_train, y_train, X_val, y_val = data

564
565
566
567
    assert hpt_type in ["bayopt", "hyperband"], (
        "Invalid hyperparameter tuning framework. " "Select one of bayopt and hyperband"
    )

lucas_miranda's avatar
lucas_miranda committed
568
    if hypermodel == "S2SAE":  # pragma: no cover
569
        assert (
570
            next_sequence_prediction == 0.0 and phenotype_prediction == 0.0
571
        ), "Prediction branches are only available for variational models. See documentation for more details"
572
        batch_size = 1
573
        hypermodel = deepof.hypermodels.SEQ_2_SEQ_AE(input_shape=X_train.shape)
574
575

    elif hypermodel == "S2SGMVAE":
576
        batch_size = 64
577
        hypermodel = deepof.hypermodels.SEQ_2_SEQ_GMVAE(
578
            input_shape=X_train.shape,
579
            encoding=encoding_size,
580
            kl_warmup_epochs=kl_warmup_epochs,
581
            loss=loss,
582
            mmd_warmup_epochs=mmd_warmup_epochs,
583
            number_of_components=k,
584
            overlap_loss=overlap_loss,
585
586
587
            next_sequence_prediction=next_sequence_prediction,
            phenotype_prediction=phenotype_prediction,
            rule_based_prediction=rule_based_prediction,
588
589
590
            rule_based_features=(
                y_train.shape[1] if not phenotype_prediction else y_train.shape[1] - 1
            ),
591
        )
lucas_miranda's avatar
lucas_miranda committed
592

593
594
595
    else:
        return False

596
597
598
    hpt_params = {
        "hypermodel": hypermodel,
        "executions_per_trial": n_replicas,
599
600
601
        "logger": TensorBoardLogger(
            metrics=["val_mae"], logdir=os.path.join(outpath, "logged_hparams")
        ),
602
603
604
605
606
607
608
        "objective": "val_mae",
        "project_name": project_name,
        "tune_new_entries": True,
    }

    if hpt_type == "hyperband":
        tuner = Hyperband(
609
610
611
            directory=os.path.join(
                outpath, "HyperBandx_{}_{}".format(loss, str(date.today()))
            ),
612
            max_epochs=30,
613
            hyperband_iterations=hypertun_trials,
614
            factor=3,
615
616
617
618
            **hpt_params
        )
    else:
        tuner = BayesianOptimization(
619
620
621
            directory=os.path.join(
                outpath, "BayOpt_{}_{}".format(loss, str(date.today()))
            ),
622
623
624
            max_trials=hypertun_trials,
            **hpt_params
        )
625
626
627

    print(tuner.search_space_summary())

628
629
630
    Xs, ys = [X_train], [X_train]
    Xvals, yvals = [X_val], [X_val]

631
    if next_sequence_prediction > 0.0:
632
633
634
        Xs, ys = X_train[:-1], [X_train[:-1], X_train[1:]]
        Xvals, yvals = X_val[:-1], [X_val[:-1], X_val[1:]]

635
    if phenotype_prediction > 0.0:
636
637
        ys += [y_train[-Xs.shape[0] :, 0]]
        yvals += [y_val[-Xvals.shape[0] :, 0]]
638
639
640
641
642
643

        # 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:
644
645
        ys += [y_train[-Xs.shape[0] :]]
        yvals += [y_val[-Xvals.shape[0] :]]
646

647
    tuner.search(
648
649
        Xs,
        ys,
650
        epochs=n_epochs,
651
        validation_data=(Xvals, yvals),
652
        verbose=1,
653
        batch_size=batch_size,
lucas_miranda's avatar
lucas_miranda committed
654
        callbacks=callbacks,
655
656
657
658
659
    )

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

lucas_miranda's avatar
lucas_miranda committed
660
661
    print(tuner.results_summary())

662
    return best_hparams, best_run