train_utils.py 17.3 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
def load_hparams(hparams):
50
51
52
53
54
55
56
57
    """Loads hyperparameters from a custom dictionary pickled on disc.
    Thought to be used with the output of hyperparameter_tuning.py"""

    if hparams is not None:
        with open(hparams, "rb") as handle:
            hparams = pickle.load(handle)
    else:
        hparams = {
58
59
60
61
62
            "bidirectional_merge": "ave",
            "clipvalue": 1.0,
            "dense_activation": "relu",
            "dense_layers_per_branch": 1,
            "dropout_rate": 1e-3,
63
            "learning_rate": 1e-3,
64
65
66
            "units_conv": 160,
            "units_dense2": 120,
            "units_lstm": 300,
67
68
69
70
71
72
73
74
75
76
77
        }
    return hparams


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,
78
                [i for i in os.listdir(train_path) if i.endswith(".pkl")][0],
79
80
81
82
83
84
85
86
87
88
89
            ),
            "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
90
91
92
93
    X_train: np.array,
    batch_size: int,
    cp: bool,
    variational: bool,
94
    phenotype_class: float,
lucas_miranda's avatar
lucas_miranda committed
95
96
    predictor: float,
    loss: str,
97
    logparam: dict = None,
98
    outpath: str = ".",
99
) -> List[Union[Any]]:
100
    """Generates callbacks for model training, including:
101
102
103
104
    - run_ID: run name, with coarse parameter details;
    - tensorboard_callback: for real-time visualization;
    - cp_callback: for checkpoint saving,
    - onecycle: for learning rate scheduling"""
105

106
    run_ID = "{}{}{}{}{}{}_{}".format(
107
        ("GMVAE" if variational else "AE"),
lucas_miranda's avatar
lucas_miranda committed
108
        ("_Pred={}".format(predictor) if predictor > 0 and variational else ""),
109
        ("_Pheno={}".format(phenotype_class) if phenotype_class > 0 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
        (datetime.now().strftime("%Y%m%d-%H%M%S")),
114
115
    )

116
    log_dir = os.path.abspath(os.path.join(outpath, "fit", run_ID))
117
    tensorboard_callback = tf.keras.callbacks.TensorBoard(
118
119
120
        log_dir=log_dir,
        histogram_freq=1,
        profile_batch=2,
121
122
123
    )

    onecycle = deepof.model_utils.one_cycle_scheduler(
124
125
        X_train.shape[0] // batch_size * 250,
        max_rate=0.005,
126
127
    )

128
129
130
131
    callbacks = [run_ID, tensorboard_callback, onecycle]

    if cp:
        cp_callback = tf.keras.callbacks.ModelCheckpoint(
132
            os.path.join(outpath, "checkpoints", run_ID + "/cp-{epoch:04d}.ckpt"),
133
134
135
136
137
138
139
140
            verbose=1,
            save_best_only=False,
            save_weights_only=True,
            save_freq="epoch",
        )
        callbacks.append(cp_callback)

    return callbacks
141
142


lucas_miranda's avatar
lucas_miranda committed
143
def log_hyperparameters(phenotype_class: float, rec: str):
lucas_miranda's avatar
lucas_miranda committed
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
192
193
194
    """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
195
196
197
198
199
200
201
202
203
204
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
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
241
242
243
    """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)


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

267
    # Load data
268
269
270
271
272
    X_train, y_train, X_val, y_val = preprocessed_object

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

273
    # Defines what to log on tensorboard (useful for trying out different models)
274
275
    logparam = {
        "encoding": encoding_size,
276
        "k": n_components,
277
278
279
280
281
        "loss": loss,
    }
    if phenotype_class:
        logparam["pheno_weight"] = phenotype_class

282
    # Load callbacks
283
    run_ID, *cbacks = get_callbacks(
284
285
286
287
        X_train=X_train,
        batch_size=batch_size,
        cp=save_checkpoints,
        variational=variational,
288
        phenotype_class=phenotype_class,
289
290
291
292
293
        predictor=predictor,
        loss=loss,
        logparam=logparam,
        outpath=output_path,
    )
294
295
    if not log_history:
        cbacks = cbacks[1:]
296

297
    # Logs hyperparameters to tensorboard
lucas_miranda's avatar
lucas_miranda committed
298
    rec = "reconstruction_" if phenotype_class else ""
299
    if log_hparams:
lucas_miranda's avatar
lucas_miranda committed
300
        logparams, metrics = log_hyperparameters(phenotype_class, rec)
301
302
303
304
305
306
307
308

        with tf.summary.create_file_writer(
            os.path.join(output_path, "hparams", run_ID)
        ).as_default():
            hp.hparams_config(
                hparams=logparams,
                metrics=metrics,
            )
309

310
    # Build models
311
312
313
314
315
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
344
    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,
        ).build(
            X_train.shape
        )
        return_list = (encoder, generator, grouper, ae)

    if pretrained:
345
        # If pretrained models are specified, load weights and return
346
347
348
349
350
351
352
353
354
        ae.load_weights(pretrained)
        return return_list

    else:
        if not variational:

            ae.fit(
                x=X_train,
                y=X_train,
355
                epochs=epochs,
356
357
358
                batch_size=batch_size,
                verbose=1,
                validation_data=(X_val, X_val),
359
360
                callbacks=cbacks
                + [
361
362
363
364
                    CustomStopper(
                        monitor="val_loss",
                        patience=5,
                        restore_best_weights=True,
365
                        start_epoch=max(kl_warmup, mmd_warmup),
366
367
368
369
                    ),
                ],
            )

370
371
372
            if save_weights:
                ae.save_weights("{}_final_weights.h5".format(run_ID))

373
374
        else:

375
            callbacks_ = cbacks + [
376
377
378
379
                CustomStopper(
                    monitor="val_loss",
                    patience=5,
                    restore_best_weights=True,
380
                    start_epoch=max(kl_warmup, mmd_warmup),
381
382
383
                ),
            ]

384
            if "ELBO" in loss and kl_warmup > 0:
385
386
                # noinspection PyUnboundLocalVariable
                callbacks_.append(kl_warmup_callback)
387
            if "MMD" in loss and mmd_warmup > 0:
388
389
390
                # noinspection PyUnboundLocalVariable
                callbacks_.append(mmd_warmup_callback)

391
392
393
394
395
396
397
            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:]]

398
            if phenotype_class > 0.0:
399
400
401
                ys += [y_train]
                yvals += [y_val]

402
            ae.fit(
403
404
                x=Xs,
                y=ys,
405
                epochs=epochs,
406
407
408
409
410
411
412
413
414
                batch_size=batch_size,
                verbose=1,
                validation_data=(
                    Xvals,
                    yvals,
                ),
                callbacks=callbacks_,
            )

415
416
417
            if save_weights:
                ae.save_weights("{}_final_weights.h5".format(run_ID))

418
419
420
421
422
            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
423
                    ae,
lucas_miranda's avatar
lucas_miranda committed
424
425
                    Xvals,
                    yvals[-1],
lucas_miranda's avatar
lucas_miranda committed
426
427
428
                    phenotype_class,
                    predictor,
                    rec,
429
                )
430

431
432
433
    return return_list


434
def tune_search(
435
    data: List[np.array],
436
    encoding_size: int,
437
438
    hypertun_trials: int,
    hpt_type: str,
439
440
    hypermodel: str,
    k: int,
441
    kl_warmup_epochs: int,
442
    loss: str,
443
    mmd_warmup_epochs: int,
444
    overlap_loss: float,
445
    phenotype_class: float,
446
447
    predictor: float,
    project_name: str,
448
    callbacks: List,
449
    n_epochs: int = 30,
450
    n_replicas: int = 1,
451
) -> Union[bool, Tuple[Any, Any]]:
452
453
    """Define the search space using keras-tuner and bayesian optimization

454
455
456
457
458
459
460
461
462
463
464
    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
465
        - phenotype_class (float): adds an extra regularizing neural network to the model,
466
467
468
469
470
471
472
473
474
475
476
477
        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
478
479
480

    """

481
482
    X_train, y_train, X_val, y_val = data

483
484
485
486
    assert hpt_type in ["bayopt", "hyperband"], (
        "Invalid hyperparameter tuning framework. " "Select one of bayopt and hyperband"
    )

lucas_miranda's avatar
lucas_miranda committed
487
    if hypermodel == "S2SAE":  # pragma: no cover
488
        assert (
489
            predictor == 0.0 and phenotype_class == 0.0
490
        ), "Prediction branches are only available for variational models. See documentation for more details"
491
        hypermodel = deepof.hypermodels.SEQ_2_SEQ_AE(input_shape=X_train.shape)
492
493
494

    elif hypermodel == "S2SGMVAE":
        hypermodel = deepof.hypermodels.SEQ_2_SEQ_GMVAE(
495
            input_shape=X_train.shape,
496
            encoding=encoding_size,
497
            kl_warmup_epochs=kl_warmup_epochs,
498
            loss=loss,
499
            mmd_warmup_epochs=mmd_warmup_epochs,
500
            number_of_components=k,
501
            overlap_loss=overlap_loss,
502
            phenotype_predictor=phenotype_class,
503
            predictor=predictor,
504
        )
lucas_miranda's avatar
lucas_miranda committed
505

506
507
508
    else:
        return False

509
510
511
512
513
514
515
516
517
518
519
520
521
522
    hpt_params = {
        "hypermodel": hypermodel,
        "executions_per_trial": n_replicas,
        "logger": TensorBoardLogger(metrics=["val_mae"], logdir="./logs/hparams"),
        "objective": "val_mae",
        "project_name": project_name,
        "seed": 42,
        "tune_new_entries": True,
    }

    if hpt_type == "hyperband":
        tuner = Hyperband(
            directory="HyperBandx_{}_{}".format(loss, str(date.today())),
            max_epochs=hypertun_trials,
523
            hyperband_iterations=3,
lucas_miranda's avatar
lucas_miranda committed
524
            factor=2,
525
526
527
528
529
530
531
532
            **hpt_params
        )
    else:
        tuner = BayesianOptimization(
            directory="BayOpt_{}_{}".format(loss, str(date.today())),
            max_trials=hypertun_trials,
            **hpt_params
        )
533
534
535

    print(tuner.search_space_summary())

536
537
538
539
540
541
542
    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:]]

543
    if phenotype_class > 0.0:
544
545
546
        ys += [y_train]
        yvals += [y_val]

547
    tuner.search(
548
549
        Xs,
        ys,
550
        epochs=n_epochs,
551
        validation_data=(Xvals, yvals),
552
553
        verbose=1,
        batch_size=256,
lucas_miranda's avatar
lucas_miranda committed
554
        callbacks=callbacks,
555
556
557
558
559
    )

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

lucas_miranda's avatar
lucas_miranda committed
560
561
    print(tuner.results_summary())

562
    return best_hparams, best_run