train_utils.py 18.2 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,
lucas_miranda's avatar
lucas_miranda committed
77
78
    knn_samples: int = 10000,
    knn_neighbors: int = 100,
79
    logparam: dict = None,
80
    outpath: str = ".",
81
) -> List[Union[Any]]:
82
    """Generates callbacks for model training, including:
83
84
85
86
    - run_ID: run name, with coarse parameter details;
    - tensorboard_callback: for real-time visualization;
    - cp_callback: for checkpoint saving,
    - onecycle: for learning rate scheduling"""
87

88
89
90
91
92
93
94
95
    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"

96
    run_ID = "{}{}{}{}{}{}{}_{}".format(
97
        ("GMVAE" if variational else "AE"),
lucas_miranda's avatar
lucas_miranda committed
98
        ("_Pred={}".format(predictor) if predictor > 0 and variational else ""),
99
        ("_Pheno={}".format(phenotype_class) if phenotype_class > 0 else ""),
100
        ("_loss={}".format(loss) if variational else ""),
101
102
        ("_encoding={}".format(logparam["encoding"]) if logparam is not None else ""),
        ("_k={}".format(logparam["k"]) if logparam is not None else ""),
103
        ("_latreg={}".format(latreg)),
104
        (datetime.now().strftime("%Y%m%d-%H%M%S")),
105
106
    )

107
    log_dir = os.path.abspath(os.path.join(outpath, "fit", run_ID))
108
    tensorboard_callback = tf.keras.callbacks.TensorBoard(
109
110
111
        log_dir=log_dir,
        histogram_freq=1,
        profile_batch=2,
112
113
    )

lucas_miranda's avatar
lucas_miranda committed
114
115
116
117
118
    knn = deepof.model_utils.knn_cluster_purity(
        k=knn_neighbors,
        samples=knn_samples,
    )

119
    onecycle = deepof.model_utils.one_cycle_scheduler(
120
121
        X_train.shape[0] // batch_size * 250,
        max_rate=0.005,
lucas_miranda's avatar
lucas_miranda committed
122
        log_dir=os.path.join(outpath, "metrics"),
123
124
    )

lucas_miranda's avatar
lucas_miranda committed
125
    callbacks = [run_ID, tensorboard_callback, knn, onecycle]
126
127
128

    if cp:
        cp_callback = tf.keras.callbacks.ModelCheckpoint(
129
            os.path.join(outpath, "checkpoints", run_ID + "/cp-{epoch:04d}.ckpt"),
130
131
132
133
134
135
136
137
            verbose=1,
            save_best_only=False,
            save_weights_only=True,
            save_freq="epoch",
        )
        callbacks.append(cp_callback)

    return callbacks
138
139


lucas_miranda's avatar
lucas_miranda committed
140
def log_hyperparameters(phenotype_class: float, rec: str):
lucas_miranda's avatar
lucas_miranda committed
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
    """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
192
193
194
195
196
197
198
199
200
201
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
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
    """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)


241
def autoencoder_fitting(
242
    preprocessed_object: Tuple[np.ndarray, np.ndarray, np.ndarray, np.ndarray],
243
244
    batch_size: int,
    encoding_size: int,
245
    epochs: int,
246
247
248
249
250
    hparams: dict,
    kl_warmup: int,
    log_history: bool,
    log_hparams: bool,
    loss: str,
251
252
253
254
255
    mmd_warmup: int,
    montecarlo_kl: int,
    n_components: int,
    output_path: str,
    phenotype_class: float,
256
257
258
    predictor: float,
    pretrained: str,
    save_checkpoints: bool,
259
    save_weights: bool,
260
    variational: bool,
261
262
    reg_cat_clusters: bool,
    reg_cluster_variance: bool,
lucas_miranda's avatar
lucas_miranda committed
263
264
    knn_neighbors: int,
    knn_samples: int,
265
):
266
267
    """Implementation function for deepof.data.coordinates.deep_unsupervised_embedding"""

268
    # Load data
269
270
271
272
273
    X_train, y_train, X_val, y_val = preprocessed_object

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

274
    # Defines what to log on tensorboard (useful for trying out different models)
275
276
    logparam = {
        "encoding": encoding_size,
277
        "k": n_components,
278
279
280
281
282
        "loss": loss,
    }
    if phenotype_class:
        logparam["pheno_weight"] = phenotype_class

283
    # Load callbacks
284
    run_ID, *cbacks = get_callbacks(
285
286
287
288
        X_train=X_train,
        batch_size=batch_size,
        cp=save_checkpoints,
        variational=variational,
289
        phenotype_class=phenotype_class,
290
291
        predictor=predictor,
        loss=loss,
lucas_miranda's avatar
lucas_miranda committed
292
293
        knn_neighbors=knn_neighbors,
        knn_samples=knn_samples,
294
        reg_cat_clusters=reg_cat_clusters,
295
        reg_cluster_variance=reg_cluster_variance,
296
297
298
        logparam=logparam,
        outpath=output_path,
    )
299
300
    if not log_history:
        cbacks = cbacks[1:]
301

302
    # Logs hyperparameters to tensorboard
lucas_miranda's avatar
lucas_miranda committed
303
    rec = "reconstruction_" if phenotype_class else ""
304
    if log_hparams:
lucas_miranda's avatar
lucas_miranda committed
305
        logparams, metrics = log_hyperparameters(phenotype_class, rec)
306
307
308
309
310
311
312
313

        with tf.summary.create_file_writer(
            os.path.join(output_path, "hparams", run_ID)
        ).as_default():
            hp.hparams_config(
                hparams=logparams,
                metrics=metrics,
            )
314

315
    # Build models
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
    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,
344
345
            reg_cat_clusters=reg_cat_clusters,
            reg_cluster_variance=reg_cluster_variance,
346
347
348
349
350
351
        ).build(
            X_train.shape
        )
        return_list = (encoder, generator, grouper, ae)

    if pretrained:
352
        # If pretrained models are specified, load weights and return
353
354
355
356
357
358
359
360
361
        ae.load_weights(pretrained)
        return return_list

    else:
        if not variational:

            ae.fit(
                x=X_train,
                y=X_train,
362
                epochs=epochs,
363
364
365
                batch_size=batch_size,
                verbose=1,
                validation_data=(X_val, X_val),
366
367
                callbacks=cbacks
                + [
368
369
370
371
                    CustomStopper(
                        monitor="val_loss",
                        patience=5,
                        restore_best_weights=True,
372
                        start_epoch=max(kl_warmup, mmd_warmup),
373
374
375
376
                    ),
                ],
            )

377
378
379
            if save_weights:
                ae.save_weights("{}_final_weights.h5".format(run_ID))

380
381
        else:

382
            callbacks_ = cbacks + [
383
384
385
386
                CustomStopper(
                    monitor="val_loss",
                    patience=5,
                    restore_best_weights=True,
387
                    start_epoch=max(kl_warmup, mmd_warmup),
388
389
390
                ),
            ]

391
            if "ELBO" in loss and kl_warmup > 0:
392
393
                # noinspection PyUnboundLocalVariable
                callbacks_.append(kl_warmup_callback)
394
            if "MMD" in loss and mmd_warmup > 0:
395
396
397
                # noinspection PyUnboundLocalVariable
                callbacks_.append(mmd_warmup_callback)

398
399
400
401
402
403
404
            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:]]

405
            if phenotype_class > 0.0:
406
407
408
                ys += [y_train]
                yvals += [y_val]

409
            ae.fit(
410
411
                x=Xs,
                y=ys,
412
                epochs=epochs,
413
414
415
416
417
418
419
420
421
                batch_size=batch_size,
                verbose=1,
                validation_data=(
                    Xvals,
                    yvals,
                ),
                callbacks=callbacks_,
            )

422
            if not os.path.exists(os.path.join(output_path, "trained_weights")):
423
424
                os.makedirs("trained_weights")

425
            if save_weights:
426
427
                ae.save_weights(
                    os.path.join(
428
429
430
                        "{}".format(output_path),
                        "trained_weights",
                        "{}_final_weights.h5".format(run_ID),
431
432
                    )
                )
433

434
435
436
437
438
            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
439
                    ae,
lucas_miranda's avatar
lucas_miranda committed
440
441
                    Xvals,
                    yvals[-1],
lucas_miranda's avatar
lucas_miranda committed
442
443
444
                    phenotype_class,
                    predictor,
                    rec,
445
                )
446

447
448
449
    return return_list


450
def tune_search(
451
    data: List[np.array],
452
    encoding_size: int,
453
454
    hypertun_trials: int,
    hpt_type: str,
455
456
    hypermodel: str,
    k: int,
457
    kl_warmup_epochs: int,
458
    loss: str,
459
    mmd_warmup_epochs: int,
460
    overlap_loss: float,
461
    phenotype_class: float,
462
463
    predictor: float,
    project_name: str,
464
    callbacks: List,
465
    n_epochs: int = 30,
466
    n_replicas: int = 1,
467
    outpath: str = ".",
468
) -> Union[bool, Tuple[Any, Any]]:
469
470
    """Define the search space using keras-tuner and bayesian optimization

471
472
473
474
475
476
477
478
479
480
481
    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
482
        - phenotype_class (float): adds an extra regularizing neural network to the model,
483
484
485
486
487
488
489
490
491
492
493
494
        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
495
496
497

    """

498
499
    X_train, y_train, X_val, y_val = data

500
501
502
503
    assert hpt_type in ["bayopt", "hyperband"], (
        "Invalid hyperparameter tuning framework. " "Select one of bayopt and hyperband"
    )

lucas_miranda's avatar
lucas_miranda committed
504
    if hypermodel == "S2SAE":  # pragma: no cover
505
        assert (
506
            predictor == 0.0 and phenotype_class == 0.0
507
        ), "Prediction branches are only available for variational models. See documentation for more details"
508
        hypermodel = deepof.hypermodels.SEQ_2_SEQ_AE(input_shape=X_train.shape)
509
510
511

    elif hypermodel == "S2SGMVAE":
        hypermodel = deepof.hypermodels.SEQ_2_SEQ_GMVAE(
512
            input_shape=X_train.shape,
513
            encoding=encoding_size,
514
            kl_warmup_epochs=kl_warmup_epochs,
515
            loss=loss,
516
            mmd_warmup_epochs=mmd_warmup_epochs,
517
            number_of_components=k,
518
            overlap_loss=overlap_loss,
519
            phenotype_predictor=phenotype_class,
520
            predictor=predictor,
521
        )
lucas_miranda's avatar
lucas_miranda committed
522

523
524
525
    else:
        return False

526
527
528
    hpt_params = {
        "hypermodel": hypermodel,
        "executions_per_trial": n_replicas,
529
530
531
        "logger": TensorBoardLogger(
            metrics=["val_mae"], logdir=os.path.join(outpath, "logged_hparams")
        ),
532
533
534
535
536
537
538
539
        "objective": "val_mae",
        "project_name": project_name,
        "seed": 42,
        "tune_new_entries": True,
    }

    if hpt_type == "hyperband":
        tuner = Hyperband(
540
541
542
            directory=os.path.join(
                outpath, "HyperBandx_{}_{}".format(loss, str(date.today()))
            ),
543
544
            max_epochs=35,
            hyperband_iterations=hypertun_trials,
545
            factor=3,
546
547
548
549
            **hpt_params
        )
    else:
        tuner = BayesianOptimization(
550
551
552
            directory=os.path.join(
                outpath, "BayOpt_{}_{}".format(loss, str(date.today()))
            ),
553
554
555
            max_trials=hypertun_trials,
            **hpt_params
        )
556
557
558

    print(tuner.search_space_summary())

559
560
561
562
563
564
565
    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:]]

566
    if phenotype_class > 0.0:
567
568
569
        ys += [y_train]
        yvals += [y_val]

570
    tuner.search(
571
572
        Xs,
        ys,
573
        epochs=n_epochs,
574
        validation_data=(Xvals, yvals),
575
        verbose=1,
576
        batch_size=64,
lucas_miranda's avatar
lucas_miranda committed
577
        callbacks=callbacks,
578
579
580
581
582
    )

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

lucas_miranda's avatar
lucas_miranda committed
583
584
    print(tuner.results_summary())

585
    return best_hparams, best_run