train_utils.py 8.44 KB
Newer Older
1
2
3
4
5
6
7
8
9
# @author lucasmiranda42
# encoding: utf-8
# module deepof

"""

Simple utility functions used in deepof example scripts. These are not part of the main package

"""
10
from datetime import date, datetime
11

12
from kerastuner import BayesianOptimization, Hyperband
13
from kerastuner import HyperParameters
14
from kerastuner_tensorboard_logger import TensorBoardLogger
15
from tensorboard.plugins.hparams import api as hp
16
from typing import Tuple, Union, Any, List
17
18
19
20
21
22
23
import deepof.hypermodels
import deepof.model_utils
import numpy as np
import os
import pickle
import tensorflow as tf

24
25
hp = HyperParameters()

26

27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
class CustomStopper(tf.keras.callbacks.EarlyStopping):
    """ Custom callback for """

    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)


46
def load_hparams(hparams):
47
48
49
50
51
52
53
54
    """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 = {
55
56
57
58
59
            "bidirectional_merge": "ave",
            "clipvalue": 1.0,
            "dense_activation": "relu",
            "dense_layers_per_branch": 1,
            "dropout_rate": 1e-3,
60
            "learning_rate": 1e-3,
61
62
63
            "units_conv": 160,
            "units_dense2": 120,
            "units_lstm": 300,
64
65
66
67
68
69
70
71
72
73
74
        }
    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,
75
                [i for i in os.listdir(train_path) if i.endswith(".pkl")][0],
76
77
78
79
80
81
82
83
84
85
86
            ),
            "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
87
88
89
90
    X_train: np.array,
    batch_size: int,
    cp: bool,
    variational: bool,
91
    phenotype_class: float,
lucas_miranda's avatar
lucas_miranda committed
92
93
    predictor: float,
    loss: str,
94
    logparam: dict = None,
95
    outpath: str = ".",
96
) -> List[Union[Any]]:
97
    """Generates callbacks for model training, including:
98
99
100
101
    - run_ID: run name, with coarse parameter details;
    - tensorboard_callback: for real-time visualization;
    - cp_callback: for checkpoint saving,
    - onecycle: for learning rate scheduling"""
102

103
    run_ID = "{}{}{}{}{}{}_{}".format(
104
        ("GMVAE" if variational else "AE"),
105
106
        ("Pred={}".format(predictor) if predictor > 0 and variational else ""),
        ("_Pheno={}".format(phenotype_class) if phenotype_class > 0 else ""),
107
        ("_loss={}".format(loss) if variational else ""),
108
109
        ("_encoding={}".format(logparam["encoding"]) if logparam is not None else ""),
        ("_k={}".format(logparam["k"]) if logparam is not None else ""),
110
        (datetime.now().strftime("%Y%m%d-%H%M%S")),
111
112
    )

113
    log_dir = os.path.abspath(os.path.join(outpath, "fit", run_ID))
114
    tensorboard_callback = tf.keras.callbacks.TensorBoard(
115
116
117
        log_dir=log_dir,
        histogram_freq=1,
        profile_batch=2,
118
119
120
    )

    onecycle = deepof.model_utils.one_cycle_scheduler(
121
122
        X_train.shape[0] // batch_size * 250,
        max_rate=0.005,
123
124
    )

125
126
127
128
    callbacks = [run_ID, tensorboard_callback, onecycle]

    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
140


def tune_search(
141
    data: List[np.array],
142
    encoding_size: int,
143
144
    hypertun_trials: int,
    hpt_type: str,
145
146
    hypermodel: str,
    k: int,
147
    kl_warmup_epochs: int,
148
    loss: str,
149
    mmd_warmup_epochs: int,
150
    overlap_loss: float,
151
    pheno_class: float,
152
153
    predictor: float,
    project_name: str,
154
    callbacks: List,
155
    n_epochs: int = 30,
156
    n_replicas: int = 1,
157
) -> Union[bool, Tuple[Any, Any]]:
158
159
    """Define the search space using keras-tuner and bayesian optimization

160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
    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
        - pheno_class (float): adds an extra regularizing neural network to the model,
        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
184
185
186

    """

187
188
    X_train, y_train, X_val, y_val = data

189
190
191
192
    assert hpt_type in ["bayopt", "hyperband"], (
        "Invalid hyperparameter tuning framework. " "Select one of bayopt and hyperband"
    )

lucas_miranda's avatar
lucas_miranda committed
193
    if hypermodel == "S2SAE":  # pragma: no cover
194
195
196
        assert (
            predictor == 0.0 and pheno_class == 0.0
        ), "Prediction branches are only available for variational models. See documentation for more details"
197
        hypermodel = deepof.hypermodels.SEQ_2_SEQ_AE(input_shape=X_train.shape)
198
199
200

    elif hypermodel == "S2SGMVAE":
        hypermodel = deepof.hypermodels.SEQ_2_SEQ_GMVAE(
201
            input_shape=X_train.shape,
202
            encoding=encoding_size,
203
            kl_warmup_epochs=kl_warmup_epochs,
204
            loss=loss,
205
            mmd_warmup_epochs=mmd_warmup_epochs,
206
            number_of_components=k,
207
            overlap_loss=overlap_loss,
208
            phenotype_predictor=pheno_class,
209
            predictor=predictor,
210
        )
lucas_miranda's avatar
lucas_miranda committed
211

212
213
214
    else:
        return False

215
216
217
218
219
220
221
222
223
224
225
226
227
228
    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,
229
            hyperband_iterations=3,
lucas_miranda's avatar
lucas_miranda committed
230
            factor=2,
231
232
233
234
235
236
237
238
            **hpt_params
        )
    else:
        tuner = BayesianOptimization(
            directory="BayOpt_{}_{}".format(loss, str(date.today())),
            max_trials=hypertun_trials,
            **hpt_params
        )
239
240
241

    print(tuner.search_space_summary())

242
243
244
245
246
247
248
249
250
251
252
    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:]]

    if pheno_class > 0.0:
        ys += [y_train]
        yvals += [y_val]

253
    tuner.search(
254
255
        Xs,
        ys,
256
        epochs=n_epochs,
257
        validation_data=(Xvals, yvals),
258
259
        verbose=1,
        batch_size=256,
lucas_miranda's avatar
lucas_miranda committed
260
        callbacks=callbacks,
261
262
263
264
265
    )

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

lucas_miranda's avatar
lucas_miranda committed
266
267
    print(tuner.results_summary())

268
    return best_hparams, best_run