train_utils.py 22.9 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
    loss: str,
76
77
    loss_warmup: int = 0,
    warmup_mode: str = "none",
78
    X_val: np.array = None,
79
    input_type: str = False,
80
81
82
83
84
85
86
    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 = ".",
87
    run: int = False,
88
) -> List[Union[Any]]:
89
    """Generates callbacks for model training, including:
90
91
    - run_ID: run name, with coarse parameter details;
    - tensorboard_callback: for real-time visualization;
92
93
94
95
    - cp_callback: for checkpoint saving;
    - onecycle: for learning rate scheduling;
    - entropy: neighborhood entropy in the latent space;
    """
96

97
98
99
100
101
102
103
104
    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"

105
    run_ID = "{}{}{}{}{}{}{}{}{}{}{}{}{}{}{}".format(
106
        ("GMVAE" if variational else "AE"),
107
        ("_input_type={}".format(input_type) if input_type else "coords"),
108
        ("_window_size={}".format(X_train.shape[1])),
109
110
111
        ("_NextSeqPred={}".format(next_sequence_prediction) if variational else ""),
        ("_PhenoPred={}".format(phenotype_prediction) if variational else ""),
        ("_RuleBasedPred={}".format(rule_based_prediction) if variational else ""),
112
        ("_loss={}".format(loss) if variational else ""),
113
114
        ("_loss_warmup={}".format(loss_warmup)),
        ("_warmup_mode={}".format(warmup_mode)),
115
116
        ("_encoding={}".format(logparam["encoding"]) if logparam is not None else ""),
        ("_k={}".format(logparam["k"]) if logparam is not None else ""),
117
        ("_latreg={}".format(latreg)),
118
119
        ("_entknn={}".format(entropy_knn)),
        ("_run={}".format(run) if run else ""),
120
        ("_{}".format(datetime.now().strftime("%Y%m%d-%H%M%S")) if not run else ""),
121
122
    )

123
    log_dir = os.path.abspath(os.path.join(outpath, "fit", run_ID))
124
    tensorboard_callback = tf.keras.callbacks.TensorBoard(
125
126
127
        log_dir=log_dir,
        histogram_freq=1,
        profile_batch=2,
128
129
    )

130
    entropy = deepof.model_utils.neighbor_latent_entropy(
131
        encoding_dim=logparam["encoding"],
132
        k=entropy_knn,
133
        samples=entropy_samples,
lucas_miranda's avatar
lucas_miranda committed
134
        validation_data=X_val,
135
        log_dir=os.path.join(outpath, "metrics", run_ID),
136
        variational=variational,
lucas_miranda's avatar
lucas_miranda committed
137
138
    )

139
    onecycle = deepof.model_utils.one_cycle_scheduler(
140
141
        X_train.shape[0] // batch_size * 250,
        max_rate=0.005,
142
        log_dir=os.path.join(outpath, "metrics", run_ID),
143
144
    )

145
    callbacks = [run_ID, tensorboard_callback, entropy, onecycle]
146
147
148

    if cp:
        cp_callback = tf.keras.callbacks.ModelCheckpoint(
149
            os.path.join(outpath, "checkpoints", run_ID + "/cp-{epoch:04d}.ckpt"),
150
151
152
153
154
155
156
157
            verbose=1,
            save_best_only=False,
            save_weights_only=True,
            save_freq="epoch",
        )
        callbacks.append(cp_callback)

    return callbacks
158
159


lucas_miranda's avatar
lucas_miranda committed
160
def log_hyperparameters(phenotype_class: float, rec: str):
lucas_miranda's avatar
lucas_miranda committed
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
    """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
212
def tensorboard_metric_logging(
213
214
215
216
217
    run_dir: str,
    hpms: Any,
    ae: Any,
    X_val: np.ndarray,
    y_val: np.ndarray,
218
219
220
    next_sequence_prediction: float,
    phenotype_prediction: float,
    rule_based_prediction: float,
221
    rec: str,
lucas_miranda's avatar
lucas_miranda committed
222
):
lucas_miranda's avatar
lucas_miranda committed
223
224
    """Autoencoder metric logging in tensorboard"""

225
226
    outputs = ae.predict(X_val)
    idx_generator = (idx for idx in range(len(outputs)))
lucas_miranda's avatar
lucas_miranda committed
227
228
229

    with tf.summary.create_file_writer(run_dir).as_default():
        hp.hparams(hpms)  # record the values used in this trial
230
231
        idx = next(idx_generator)

lucas_miranda's avatar
lucas_miranda committed
232
        val_mae = tf.reduce_mean(
233
            tf.keras.metrics.mean_absolute_error(y_val[idx], outputs[idx])
lucas_miranda's avatar
lucas_miranda committed
234
235
        )
        val_mse = tf.reduce_mean(
236
            tf.keras.metrics.mean_squared_error(y_val[idx], outputs[idx])
lucas_miranda's avatar
lucas_miranda committed
237
238
239
240
        )
        tf.summary.scalar("val_{}mae".format(rec), val_mae, step=1)
        tf.summary.scalar("val_{}mse".format(rec), val_mse, step=1)

241
        if next_sequence_prediction:
242
            idx = next(idx_generator)
lucas_miranda's avatar
lucas_miranda committed
243
            pred_mae = tf.reduce_mean(
244
                tf.keras.metrics.mean_absolute_error(y_val[idx], outputs[idx])
lucas_miranda's avatar
lucas_miranda committed
245
246
            )
            pred_mse = tf.reduce_mean(
247
                tf.keras.metrics.mean_squared_error(y_val[idx], outputs[idx])
248
249
250
251
252
253
            )
            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
254
255
            )

256
        if phenotype_prediction:
257
            idx = next(idx_generator)
258
259
260
            pheno_acc = tf.keras.metrics.binary_accuracy(
                y_val[idx], tf.squeeze(outputs[idx])
            )
261
            pheno_auc = tf.keras.metrics.AUC()
262
            pheno_auc.update_state(y_val[idx], outputs[idx])
263
            pheno_auc = pheno_auc.result().numpy()
lucas_miranda's avatar
lucas_miranda committed
264
265
266
267

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

268
        if rule_based_prediction:
269
            idx = next(idx_generator)
270
            rules_mae = tf.reduce_mean(
271
                tf.keras.metrics.mean_absolute_error(y_val[idx], outputs[idx])
272
273
            )
            rules_mse = tf.reduce_mean(
274
                tf.keras.metrics.mean_squared_error(y_val[idx], outputs[idx])
275
276
277
278
            )
            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
279

280
def autoencoder_fitting(
281
282
283
284
285
    preprocessed_object: Tuple[np.ndarray, np.ndarray, np.ndarray, np.ndarray],
    batch_size: int,
    encoding_size: int,
    epochs: int,
    hparams: dict,
286
    kl_annealing_mode: str,
287
288
289
290
    kl_warmup: int,
    log_history: bool,
    log_hparams: bool,
    loss: str,
291
    mmd_annealing_mode: str,
292
293
294
295
    mmd_warmup: int,
    montecarlo_kl: int,
    n_components: int,
    output_path: str,
296
297
298
    next_sequence_prediction: float,
    phenotype_prediction: float,
    rule_based_prediction: float,
299
300
301
302
303
304
305
306
    pretrained: str,
    save_checkpoints: bool,
    save_weights: bool,
    variational: bool,
    reg_cat_clusters: bool,
    reg_cluster_variance: bool,
    entropy_samples: int,
    entropy_knn: int,
307
    input_type: str,
308
    run: int = 0,
309
    strategy: tf.distribute.Strategy = tf.distribute.MirroredStrategy(),
310
):
311
312
    """Implementation function for deepof.data.coordinates.deep_unsupervised_embedding"""

313
    # Load data
314
315
316
317
318
    X_train, y_train, X_val, y_val = preprocessed_object

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

319
320
321
322
323
324
325
326
327
328
329
330
331
    # Set options for tf.data.Datasets
    options = tf.data.Options()
    options.experimental_distribute.auto_shard_policy = (
        tf.data.experimental.AutoShardPolicy.DATA
    )

    # Generate validation dataset for callback usage
    X_val_dataset = (
        tf.data.Dataset.from_tensor_slices(X_val)
        .with_options(options)
        .batch(batch_size)
    )

332
    # Defines what to log on tensorboard (useful for trying out different models)
333
334
    logparam = {
        "encoding": encoding_size,
335
        "k": n_components,
336
337
        "loss": loss,
    }
338
339
    if phenotype_prediction:
        logparam["pheno_weight"] = phenotype_prediction
340

341
    # Load callbacks
342
    run_ID, *cbacks = get_callbacks(
343
344
345
        X_train=X_train,
        batch_size=batch_size,
        variational=variational,
346
        phenotype_prediction=phenotype_prediction,
347
        next_sequence_prediction=next_sequence_prediction,
348
        rule_based_prediction=rule_based_prediction,
349
        loss=loss,
350
351
        loss_warmup=kl_warmup,
        warmup_mode=kl_annealing_mode,
352
        input_type=input_type,
353
        X_val=(X_val_dataset if X_val.shape != (0,) else None),
354
        cp=save_checkpoints,
355
        reg_cat_clusters=reg_cat_clusters,
356
        reg_cluster_variance=reg_cluster_variance,
357
358
        entropy_samples=entropy_samples,
        entropy_knn=entropy_knn,
359
360
        logparam=logparam,
        outpath=output_path,
361
        run=run,
362
    )
363
364
    if not log_history:
        cbacks = cbacks[1:]
365

366
    # Logs hyperparameters to tensorboard
367
    rec = "reconstruction_" if phenotype_prediction else ""
368
    if log_hparams:
369
        logparams, metrics = log_hyperparameters(phenotype_prediction, rec)
370
371

        with tf.summary.create_file_writer(
372
            os.path.join(output_path, "hparams", run_ID)
373
374
375
376
377
        ).as_default():
            hp.hparams_config(
                hparams=logparams,
                metrics=metrics,
            )
378

379
380
381
382
383
384
385
386
    # 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

387
    # Build models
388
389
390
391
392
393
394
    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:
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
        with strategy.scope():
            (
                encoder,
                generator,
                grouper,
                ae,
                prior,
                posterior,
            ) = deepof.models.SEQ_2_SEQ_GMVAE(
                architecture_hparams=({} if hparams is None else hparams),
                batch_size=batch_size * strategy.num_replicas_in_sync,
                compile_model=True,
                encoding=encoding_size,
                kl_annealing_mode=kl_annealing_mode,
                kl_warmup_epochs=kl_warmup,
                loss=loss,
                mmd_annealing_mode=mmd_annealing_mode,
                mmd_warmup_epochs=mmd_warmup,
                montecarlo_kl=montecarlo_kl,
                neuron_control=False,
                number_of_components=n_components,
                overlap_loss=False,
                next_sequence_prediction=next_sequence_prediction,
                phenotype_prediction=phenotype_prediction,
                rule_based_prediction=rule_based_prediction,
                rule_based_features=rule_based_features,
                reg_cat_clusters=reg_cat_clusters,
                reg_cluster_variance=reg_cluster_variance,
            ).build(
                X_train.shape
            )
            return_list = (encoder, generator, grouper, ae)
427
428

    if pretrained:
429
        # If pretrained models are specified, load weights and return
430
431
432
433
434
435
436
437
438
        ae.load_weights(pretrained)
        return return_list

    else:
        if not variational:

            ae.fit(
                x=X_train,
                y=X_train,
439
                epochs=epochs,
440
441
                verbose=1,
                validation_data=(X_val, X_val),
442
                callbacks=cbacks
443
444
445
446
447
448
449
450
                + [
                    CustomStopper(
                        monitor="val_loss",
                        patience=5,
                        restore_best_weights=True,
                        start_epoch=max(kl_warmup, mmd_warmup),
                    ),
                ],
451
452
            )

453
454
455
            if not os.path.exists(os.path.join(output_path, "trained_weights")):
                os.makedirs(os.path.join(output_path, "trained_weights"))

456
            if save_weights:
457
458
459
460
461
462
463
                ae.save_weights(
                    os.path.join(
                        "{}".format(output_path),
                        "trained_weights",
                        "{}_final_weights.h5".format(run_ID),
                    )
                )
464

465
466
        else:

467
            callbacks_ = cbacks + [
468
469
470
471
                CustomStopper(
                    monitor="val_loss",
                    patience=5,
                    restore_best_weights=True,
472
                    start_epoch=max(kl_warmup, mmd_warmup),
473
474
475
                ),
            ]

476
477
            Xs, ys = X_train, [X_train]
            Xvals, yvals = X_val, [X_val]
478

479
            if next_sequence_prediction > 0.0:
480
481
482
                Xs, ys = X_train[:-1], [X_train[:-1], X_train[1:]]
                Xvals, yvals = X_val[:-1], [X_val[:-1], X_val[1:]]

483
            if phenotype_prediction > 0.0:
484
                ys += [y_train[-Xs.shape[0] :, 0]]
485
                yvals += [y_val[-Xvals.shape[0] :, 0]]
486
487
488
489
490
491

                # 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:
492
                ys += [y_train[-Xs.shape[0] :]]
493
                yvals += [y_val[-Xvals.shape[0] :]]
494

495
496
            # Convert data to tf.data.Dataset objects
            train_dataset = (
497
498
499
                tf.data.Dataset.from_tensor_slices((Xs, tuple(ys)))
                .batch(batch_size * strategy.num_replicas_in_sync)
                .shuffle(buffer_size=X_train.shape[0])
500
501
502
                .with_options(options)
            )
            val_dataset = (
503
504
                tf.data.Dataset.from_tensor_slices((Xvals, tuple(yvals)))
                .batch(batch_size * strategy.num_replicas_in_sync)
505
506
507
                .with_options(options)
            )

508
            ae.fit(
509
                x=train_dataset,
510
                epochs=epochs,
511
                verbose=1,
512
                validation_data=val_dataset,
513
514
515
                callbacks=callbacks_,
            )

516
            if not os.path.exists(os.path.join(output_path, "trained_weights")):
517
                os.makedirs(os.path.join(output_path, "trained_weights"))
518

519
            if save_weights:
520
521
                ae.save_weights(
                    os.path.join(
522
523
524
                        "{}".format(output_path),
                        "trained_weights",
                        "{}_final_weights.h5".format(run_ID),
525
526
                    )
                )
527

528
529
530
            if log_hparams:
                # Logparams to tensorboard
                tensorboard_metric_logging(
531
532
533
534
535
536
537
538
539
                    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,
540
                )
541

542
543
544
    return return_list


545
def tune_search(
546
547
548
549
550
551
552
553
554
555
    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,
556
557
558
    next_sequence_prediction: float,
    phenotype_prediction: float,
    rule_based_prediction: float,
559
560
561
562
563
    project_name: str,
    callbacks: List,
    n_epochs: int = 30,
    n_replicas: int = 1,
    outpath: str = ".",
564
) -> Union[bool, Tuple[Any, Any]]:
565
566
    """Define the search space using keras-tuner and bayesian optimization

567
568
569
570
571
572
573
574
575
576
577
    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
578
        - phenotype_class (float): adds an extra regularizing neural network to the model,
579
580
581
582
583
584
585
586
587
588
589
590
        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
591
592
593

    """

594
595
    X_train, y_train, X_val, y_val = data

596
597
598
599
    assert hpt_type in ["bayopt", "hyperband"], (
        "Invalid hyperparameter tuning framework. " "Select one of bayopt and hyperband"
    )

lucas_miranda's avatar
lucas_miranda committed
600
    if hypermodel == "S2SAE":  # pragma: no cover
601
        assert (
602
            next_sequence_prediction == 0.0 and phenotype_prediction == 0.0
603
        ), "Prediction branches are only available for variational models. See documentation for more details"
604
        batch_size = 1
605
        hypermodel = deepof.hypermodels.SEQ_2_SEQ_AE(input_shape=X_train.shape)
606
607

    elif hypermodel == "S2SGMVAE":
608
        batch_size = 64
609
        hypermodel = deepof.hypermodels.SEQ_2_SEQ_GMVAE(
610
            input_shape=X_train.shape,
611
            encoding=encoding_size,
612
            kl_warmup_epochs=kl_warmup_epochs,
613
            loss=loss,
614
            mmd_warmup_epochs=mmd_warmup_epochs,
615
            number_of_components=k,
616
            overlap_loss=overlap_loss,
617
618
619
            next_sequence_prediction=next_sequence_prediction,
            phenotype_prediction=phenotype_prediction,
            rule_based_prediction=rule_based_prediction,
620
621
622
            rule_based_features=(
                y_train.shape[1] if not phenotype_prediction else y_train.shape[1] - 1
            ),
623
        )
lucas_miranda's avatar
lucas_miranda committed
624

625
626
627
    else:
        return False

628
629
630
    hpt_params = {
        "hypermodel": hypermodel,
        "executions_per_trial": n_replicas,
631
632
633
        "logger": TensorBoardLogger(
            metrics=["val_mae"], logdir=os.path.join(outpath, "logged_hparams")
        ),
634
635
636
637
638
639
640
        "objective": "val_mae",
        "project_name": project_name,
        "tune_new_entries": True,
    }

    if hpt_type == "hyperband":
        tuner = Hyperband(
641
642
643
            directory=os.path.join(
                outpath, "HyperBandx_{}_{}".format(loss, str(date.today()))
            ),
644
            max_epochs=30,
645
            hyperband_iterations=hypertun_trials,
646
            factor=3,
647
648
649
650
            **hpt_params
        )
    else:
        tuner = BayesianOptimization(
651
652
653
            directory=os.path.join(
                outpath, "BayOpt_{}_{}".format(loss, str(date.today()))
            ),
654
655
656
            max_trials=hypertun_trials,
            **hpt_params
        )
657
658
659

    print(tuner.search_space_summary())

660
661
662
    Xs, ys = [X_train], [X_train]
    Xvals, yvals = [X_val], [X_val]

663
    if next_sequence_prediction > 0.0:
664
665
666
        Xs, ys = X_train[:-1], [X_train[:-1], X_train[1:]]
        Xvals, yvals = X_val[:-1], [X_val[:-1], X_val[1:]]

667
    if phenotype_prediction > 0.0:
668
669
        ys += [y_train[-Xs.shape[0] :, 0]]
        yvals += [y_val[-Xvals.shape[0] :, 0]]
670
671
672
673
674
675

        # 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:
676
677
        ys += [y_train[-Xs.shape[0] :]]
        yvals += [y_val[-Xvals.shape[0] :]]
678

679
    tuner.search(
680
681
        Xs,
        ys,
682
        epochs=n_epochs,
683
        validation_data=(Xvals, yvals),
684
        verbose=1,
685
        batch_size=batch_size,
lucas_miranda's avatar
lucas_miranda committed
686
        callbacks=callbacks,
687
688
689
690
691
    )

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

lucas_miranda's avatar
lucas_miranda committed
692
693
    print(tuner.results_summary())

694
    return best_hparams, best_run