train_utils.py 17.8 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

"""

11
from datetime import date, datetime
12
from kerastuner import BayesianOptimization, Hyperband
13
from kerastuner import HyperParameters
14
from kerastuner_tensorboard_logger import TensorBoardLogger
lucas_miranda's avatar
lucas_miranda committed
15
from sklearn.metrics import roc_auc_score
16
from tensorboard.plugins.hparams import api as hp
17
from typing import Tuple, Union, Any, List
18
19
20
21
22
23
24
import deepof.hypermodels
import deepof.model_utils
import numpy as np
import os
import pickle
import tensorflow as tf

25
26
27
28
29
# Ignore warning with no downstream effect
tf.get_logger().setLevel("ERROR")
tf.autograph.set_verbosity(0)


30
class CustomStopper(tf.keras.callbacks.EarlyStopping):
31
    """ Custom early stopping callback. Prevents the model from stopping before warmup is over """
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48

    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)


49
50
51
52
53
54
55
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(
            os.path.join(
                train_path,
56
                [i for i in os.listdir(train_path) if i.endswith(".pkl")][0],
57
58
59
60
61
62
63
64
65
66
67
            ),
            "rb",
        ) as handle:
            treatment_dict = pickle.load(handle)
    except IndexError:
        treatment_dict = None

    return treatment_dict


def get_callbacks(
lucas_miranda's avatar
lucas_miranda committed
68
69
70
    X_train: np.array,
    batch_size: int,
    variational: bool,
71
    phenotype_class: float,
lucas_miranda's avatar
lucas_miranda committed
72
73
    predictor: float,
    loss: str,
74
75
76
    cp: bool = False,
    reg_cat_clusters: bool = False,
    reg_cluster_variance: bool = False,
77
    logparam: dict = None,
78
    outpath: str = ".",
79
) -> List[Union[Any]]:
80
    """Generates callbacks for model training, including:
81
82
83
84
    - run_ID: run name, with coarse parameter details;
    - tensorboard_callback: for real-time visualization;
    - cp_callback: for checkpoint saving,
    - onecycle: for learning rate scheduling"""
85

86
87
88
89
90
91
92
93
    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"

94
    run_ID = "{}{}{}{}{}{}{}_{}".format(
95
        ("GMVAE" if variational else "AE"),
lucas_miranda's avatar
lucas_miranda committed
96
        ("_Pred={}".format(predictor) if predictor > 0 and variational else ""),
97
        ("_Pheno={}".format(phenotype_class) if phenotype_class > 0 else ""),
98
        ("_loss={}".format(loss) if variational else ""),
99
100
        ("_encoding={}".format(logparam["encoding"]) if logparam is not None else ""),
        ("_k={}".format(logparam["k"]) if logparam is not None else ""),
101
        ("_latreg={}".format(latreg)),
102
        (datetime.now().strftime("%Y%m%d-%H%M%S")),
103
104
    )

105
    log_dir = os.path.abspath(os.path.join(outpath, "fit", run_ID))
106
    tensorboard_callback = tf.keras.callbacks.TensorBoard(
107
108
109
        log_dir=log_dir,
        histogram_freq=1,
        profile_batch=2,
110
111
112
    )

    onecycle = deepof.model_utils.one_cycle_scheduler(
113
114
        X_train.shape[0] // batch_size * 250,
        max_rate=0.005,
115
116
    )

117
118
119
120
    callbacks = [run_ID, tensorboard_callback, onecycle]

    if cp:
        cp_callback = tf.keras.callbacks.ModelCheckpoint(
121
            os.path.join(outpath, "checkpoints", run_ID + "/cp-{epoch:04d}.ckpt"),
122
123
124
125
126
127
128
129
            verbose=1,
            save_best_only=False,
            save_weights_only=True,
            save_freq="epoch",
        )
        callbacks.append(cp_callback)

    return callbacks
130
131


lucas_miranda's avatar
lucas_miranda committed
132
def log_hyperparameters(phenotype_class: float, rec: str):
lucas_miranda's avatar
lucas_miranda committed
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
    """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
184
185
186
187
188
189
190
191
192
193
def tensorboard_metric_logging(
    run_dir: str,
    hpms: Any,
    ae: Any,
    X_val: np.ndarray,
    y_val: np.ndarray,
    phenotype_class: float,
    predictor: float,
    rec: str,
):
lucas_miranda's avatar
lucas_miranda committed
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
    """Autoencoder metric logging in tensorboard"""

    output = ae.predict(X_val)
    if phenotype_class or predictor:
        reconstruction = output[0]
        prediction = output[1]
        pheno = output[-1]
    else:
        reconstruction = output

    with tf.summary.create_file_writer(run_dir).as_default():
        hp.hparams(hpms)  # record the values used in this trial
        val_mae = tf.reduce_mean(
            tf.keras.metrics.mean_absolute_error(X_val, reconstruction)
        )
        val_mse = tf.reduce_mean(
            tf.keras.metrics.mean_squared_error(X_val, reconstruction)
        )
        tf.summary.scalar("val_{}mae".format(rec), val_mae, step=1)
        tf.summary.scalar("val_{}mse".format(rec), val_mse, step=1)

        if predictor:
            pred_mae = tf.reduce_mean(
                tf.keras.metrics.mean_absolute_error(X_val, prediction)
            )
            pred_mse = tf.reduce_mean(
                tf.keras.metrics.mean_squared_error(X_val, prediction)
            )
            tf.summary.scalar("val_prediction_mae".format(rec), pred_mae, step=1)
            tf.summary.scalar("val_prediction_mse".format(rec), pred_mse, step=1)

        if phenotype_class:
            pheno_acc = tf.keras.metrics.binary_accuracy(y_val, tf.squeeze(pheno))
            pheno_auc = roc_auc_score(y_val, pheno)

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


233
def autoencoder_fitting(
234
    preprocessed_object: Tuple[np.ndarray, np.ndarray, np.ndarray, np.ndarray],
235
236
    batch_size: int,
    encoding_size: int,
237
    epochs: int,
238
239
240
241
242
    hparams: dict,
    kl_warmup: int,
    log_history: bool,
    log_hparams: bool,
    loss: str,
243
244
245
246
247
    mmd_warmup: int,
    montecarlo_kl: int,
    n_components: int,
    output_path: str,
    phenotype_class: float,
248
249
250
    predictor: float,
    pretrained: str,
    save_checkpoints: bool,
251
    save_weights: bool,
252
    variational: bool,
253
254
    reg_cat_clusters: bool,
    reg_cluster_variance: bool,
255
):
256
257
    """Implementation function for deepof.data.coordinates.deep_unsupervised_embedding"""

258
    # Load data
259
260
261
262
263
    X_train, y_train, X_val, y_val = preprocessed_object

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

264
    # Defines what to log on tensorboard (useful for trying out different models)
265
266
    logparam = {
        "encoding": encoding_size,
267
        "k": n_components,
268
269
270
271
272
        "loss": loss,
    }
    if phenotype_class:
        logparam["pheno_weight"] = phenotype_class

273
    # Load callbacks
274
    run_ID, *cbacks = get_callbacks(
275
276
277
278
        X_train=X_train,
        batch_size=batch_size,
        cp=save_checkpoints,
        variational=variational,
279
        phenotype_class=phenotype_class,
280
281
        predictor=predictor,
        loss=loss,
282
        reg_cat_clusters=reg_cat_clusters,
283
        reg_cluster_variance=reg_cluster_variance,
284
285
286
        logparam=logparam,
        outpath=output_path,
    )
287
288
    if not log_history:
        cbacks = cbacks[1:]
289

290
    # Logs hyperparameters to tensorboard
lucas_miranda's avatar
lucas_miranda committed
291
    rec = "reconstruction_" if phenotype_class else ""
292
    if log_hparams:
lucas_miranda's avatar
lucas_miranda committed
293
        logparams, metrics = log_hyperparameters(phenotype_class, rec)
294
295
296
297
298
299
300
301

        with tf.summary.create_file_writer(
            os.path.join(output_path, "hparams", run_ID)
        ).as_default():
            hp.hparams_config(
                hparams=logparams,
                metrics=metrics,
            )
302

303
    # Build models
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
    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:
        (
            encoder,
            generator,
            grouper,
            ae,
            kl_warmup_callback,
            mmd_warmup_callback,
        ) = deepof.models.SEQ_2_SEQ_GMVAE(
            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,
            phenotype_prediction=phenotype_class,
            predictor=predictor,
332
333
            reg_cat_clusters=reg_cat_clusters,
            reg_cluster_variance=reg_cluster_variance,
334
335
336
337
338
339
        ).build(
            X_train.shape
        )
        return_list = (encoder, generator, grouper, ae)

    if pretrained:
340
        # If pretrained models are specified, load weights and return
341
342
343
344
345
346
347
348
349
        ae.load_weights(pretrained)
        return return_list

    else:
        if not variational:

            ae.fit(
                x=X_train,
                y=X_train,
350
                epochs=epochs,
351
352
353
                batch_size=batch_size,
                verbose=1,
                validation_data=(X_val, X_val),
354
355
                callbacks=cbacks
                + [
356
357
358
359
                    CustomStopper(
                        monitor="val_loss",
                        patience=5,
                        restore_best_weights=True,
360
                        start_epoch=max(kl_warmup, mmd_warmup),
361
362
363
364
                    ),
                ],
            )

365
366
367
            if save_weights:
                ae.save_weights("{}_final_weights.h5".format(run_ID))

368
369
        else:

370
            callbacks_ = cbacks + [
371
372
373
374
                CustomStopper(
                    monitor="val_loss",
                    patience=5,
                    restore_best_weights=True,
375
                    start_epoch=max(kl_warmup, mmd_warmup),
376
377
378
                ),
            ]

379
            if "ELBO" in loss and kl_warmup > 0:
380
381
                # noinspection PyUnboundLocalVariable
                callbacks_.append(kl_warmup_callback)
382
            if "MMD" in loss and mmd_warmup > 0:
383
384
385
                # noinspection PyUnboundLocalVariable
                callbacks_.append(mmd_warmup_callback)

386
387
388
389
390
391
392
            Xs, ys = [X_train], [X_train]
            Xvals, yvals = [X_val], [X_val]

            if predictor > 0.0:
                Xs, ys = X_train[:-1], [X_train[:-1], X_train[1:]]
                Xvals, yvals = X_val[:-1], [X_val[:-1], X_val[1:]]

393
            if phenotype_class > 0.0:
394
395
396
                ys += [y_train]
                yvals += [y_val]

397
            ae.fit(
398
399
                x=Xs,
                y=ys,
400
                epochs=epochs,
401
402
403
404
405
406
407
408
409
                batch_size=batch_size,
                verbose=1,
                validation_data=(
                    Xvals,
                    yvals,
                ),
                callbacks=callbacks_,
            )

410
            if not os.path.exists(os.path.join(output_path, "trained_weights")):
411
412
                os.makedirs("trained_weights")

413
            if save_weights:
414
415
                ae.save_weights(
                    os.path.join(
416
417
418
                        "{}".format(output_path),
                        "trained_weights",
                        "{}_final_weights.h5".format(run_ID),
419
420
                    )
                )
421

422
423
424
425
426
            if log_hparams:
                # Logparams to tensorboard
                tensorboard_metric_logging(
                    os.path.join(output_path, "hparams", run_ID),
                    logparam,
lucas_miranda's avatar
lucas_miranda committed
427
                    ae,
lucas_miranda's avatar
lucas_miranda committed
428
429
                    Xvals,
                    yvals[-1],
lucas_miranda's avatar
lucas_miranda committed
430
431
432
                    phenotype_class,
                    predictor,
                    rec,
433
                )
434

435
436
437
    return return_list


438
def tune_search(
439
    data: List[np.array],
440
    encoding_size: int,
441
442
    hypertun_trials: int,
    hpt_type: str,
443
444
    hypermodel: str,
    k: int,
445
    kl_warmup_epochs: int,
446
    loss: str,
447
    mmd_warmup_epochs: int,
448
    overlap_loss: float,
449
    phenotype_class: float,
450
451
    predictor: float,
    project_name: str,
452
    callbacks: List,
453
    n_epochs: int = 30,
454
    n_replicas: int = 1,
455
    outpath: str = ".",
456
) -> Union[bool, Tuple[Any, Any]]:
457
458
    """Define the search space using keras-tuner and bayesian optimization

459
460
461
462
463
464
465
466
467
468
469
    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
470
        - phenotype_class (float): adds an extra regularizing neural network to the model,
471
472
473
474
475
476
477
478
479
480
481
482
        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
483
484
485

    """

486
487
    X_train, y_train, X_val, y_val = data

488
489
490
491
    assert hpt_type in ["bayopt", "hyperband"], (
        "Invalid hyperparameter tuning framework. " "Select one of bayopt and hyperband"
    )

lucas_miranda's avatar
lucas_miranda committed
492
    if hypermodel == "S2SAE":  # pragma: no cover
493
        assert (
494
            predictor == 0.0 and phenotype_class == 0.0
495
        ), "Prediction branches are only available for variational models. See documentation for more details"
496
        hypermodel = deepof.hypermodels.SEQ_2_SEQ_AE(input_shape=X_train.shape)
497
498
499

    elif hypermodel == "S2SGMVAE":
        hypermodel = deepof.hypermodels.SEQ_2_SEQ_GMVAE(
500
            input_shape=X_train.shape,
501
            encoding=encoding_size,
502
            kl_warmup_epochs=kl_warmup_epochs,
503
            loss=loss,
504
            mmd_warmup_epochs=mmd_warmup_epochs,
505
            number_of_components=k,
506
            overlap_loss=overlap_loss,
507
            phenotype_predictor=phenotype_class,
508
            predictor=predictor,
509
        )
lucas_miranda's avatar
lucas_miranda committed
510

511
512
513
    else:
        return False

514
515
516
    hpt_params = {
        "hypermodel": hypermodel,
        "executions_per_trial": n_replicas,
517
518
519
        "logger": TensorBoardLogger(
            metrics=["val_mae"], logdir=os.path.join(outpath, "logged_hparams")
        ),
520
521
522
523
524
525
526
527
        "objective": "val_mae",
        "project_name": project_name,
        "seed": 42,
        "tune_new_entries": True,
    }

    if hpt_type == "hyperband":
        tuner = Hyperband(
528
529
530
            directory=os.path.join(
                outpath, "HyperBandx_{}_{}".format(loss, str(date.today()))
            ),
531
532
            max_epochs=35,
            hyperband_iterations=hypertun_trials,
533
            factor=3,
534
535
536
537
            **hpt_params
        )
    else:
        tuner = BayesianOptimization(
538
539
540
            directory=os.path.join(
                outpath, "BayOpt_{}_{}".format(loss, str(date.today()))
            ),
541
542
543
            max_trials=hypertun_trials,
            **hpt_params
        )
544
545
546

    print(tuner.search_space_summary())

547
548
549
550
551
552
553
    Xs, ys = [X_train], [X_train]
    Xvals, yvals = [X_val], [X_val]

    if predictor > 0.0:
        Xs, ys = X_train[:-1], [X_train[:-1], X_train[1:]]
        Xvals, yvals = X_val[:-1], [X_val[:-1], X_val[1:]]

554
    if phenotype_class > 0.0:
555
556
557
        ys += [y_train]
        yvals += [y_val]

558
    tuner.search(
559
560
        Xs,
        ys,
561
        epochs=n_epochs,
562
        validation_data=(Xvals, yvals),
563
        verbose=1,
564
        batch_size=64,
lucas_miranda's avatar
lucas_miranda committed
565
        callbacks=callbacks,
566
567
568
569
570
    )

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

lucas_miranda's avatar
lucas_miranda committed
571
572
    print(tuner.results_summary())

573
    return best_hparams, best_run