train_utils.py 8.28 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 typing import Tuple, Union, Any, List
16
17
18
19
20
21
22
import deepof.hypermodels
import deepof.model_utils
import numpy as np
import os
import pickle
import tensorflow as tf

23
24
hp = HyperParameters()

25

26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
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)


45
def load_hparams(hparams):
46
47
48
49
50
51
52
53
    """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 = {
54
55
56
57
58
            "bidirectional_merge": "ave",
            "clipvalue": 1.0,
            "dense_activation": "relu",
            "dense_layers_per_branch": 1,
            "dropout_rate": 1e-3,
59
            "learning_rate": 1e-3,
60
61
62
            "units_conv": 160,
            "units_dense2": 120,
            "units_lstm": 300,
63
64
65
66
67
68
69
70
71
72
73
        }
    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,
74
                [i for i in os.listdir(train_path) if i.endswith(".pkl")][0],
75
76
77
78
79
80
81
82
83
84
85
            ),
            "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
86
87
88
89
    X_train: np.array,
    batch_size: int,
    cp: bool,
    variational: bool,
90
    phenotype_class: float,
lucas_miranda's avatar
lucas_miranda committed
91
92
    predictor: float,
    loss: str,
93
    logparam: dict = None,
94
    outpath: str = ".",
95
) -> List[Union[Any]]:
96
    """Generates callbacks for model training, including:
97
98
99
100
    - run_ID: run name, with coarse parameter details;
    - tensorboard_callback: for real-time visualization;
    - cp_callback: for checkpoint saving,
    - onecycle: for learning rate scheduling"""
101

102
    run_ID = "{}{}{}{}{}{}_{}".format(
103
104
        ("GMVAE" if variational else "AE"),
        ("P" if predictor > 0 and variational else ""),
105
        ("_Pheno" if phenotype_class > 0 else ""),
106
        ("_loss={}".format(loss) if variational else ""),
107
108
109
        ("_encoding={}".format(logparam["encoding"])),
        ("_k={}".format(logparam["k"])),
        (datetime.now().strftime("%Y%m%d-%H%M%S")),
110
111
    )

112
    log_dir = os.path.abspath(os.path.join(outpath, "fit", run_ID))
113
    tensorboard_callback = tf.keras.callbacks.TensorBoard(
114
115
116
        log_dir=log_dir,
        histogram_freq=1,
        profile_batch=2,
117
118
119
    )

    onecycle = deepof.model_utils.one_cycle_scheduler(
120
121
        X_train.shape[0] // batch_size * 250,
        max_rate=0.005,
122
123
    )

124
125
126
127
    callbacks = [run_ID, tensorboard_callback, onecycle]

    if cp:
        cp_callback = tf.keras.callbacks.ModelCheckpoint(
128
            os.path.join(outpath, "checkpoints", run_ID + "/cp-{epoch:04d}.ckpt"),
129
130
131
132
133
134
135
136
            verbose=1,
            save_best_only=False,
            save_weights_only=True,
            save_freq="epoch",
        )
        callbacks.append(cp_callback)

    return callbacks
137
138
139


def tune_search(
140
    data: List[np.array],
141
    encoding_size: int,
142
143
    hypertun_trials: int,
    hpt_type: str,
144
145
    hypermodel: str,
    k: int,
146
    kl_warmup_epochs: int,
147
    loss: str,
148
    mmd_warmup_epochs: int,
149
    overlap_loss: float,
150
    pheno_class: float,
151
152
    predictor: float,
    project_name: str,
153
    callbacks: List,
154
    n_epochs: int = 30,
155
    n_replicas: int = 1,
156
) -> Union[bool, Tuple[Any, Any]]:
157
158
    """Define the search space using keras-tuner and bayesian optimization

159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
    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
183
184
185

    """

186
187
    X_train, y_train, X_val, y_val = data

188
189
190
191
    assert hpt_type in ["bayopt", "hyperband"], (
        "Invalid hyperparameter tuning framework. " "Select one of bayopt and hyperband"
    )

lucas_miranda's avatar
lucas_miranda committed
192
    if hypermodel == "S2SAE":  # pragma: no cover
193
194
195
        assert (
            predictor == 0.0 and pheno_class == 0.0
        ), "Prediction branches are only available for variational models. See documentation for more details"
196
        hypermodel = deepof.hypermodels.SEQ_2_SEQ_AE(input_shape=X_train.shape)
197
198
199

    elif hypermodel == "S2SGMVAE":
        hypermodel = deepof.hypermodels.SEQ_2_SEQ_GMVAE(
200
            input_shape=X_train.shape,
201
            encoding=encoding_size,
202
            kl_warmup_epochs=kl_warmup_epochs,
203
            loss=loss,
204
            mmd_warmup_epochs=mmd_warmup_epochs,
205
            number_of_components=k,
206
            overlap_loss=overlap_loss,
207
            phenotype_predictor=pheno_class,
208
            predictor=predictor,
209
        )
lucas_miranda's avatar
lucas_miranda committed
210

211
212
213
    else:
        return False

214
215
216
217
218
219
220
221
222
223
224
225
226
227
    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,
228
            hyperband_iterations=3,
lucas_miranda's avatar
lucas_miranda committed
229
            factor=2,
230
231
232
233
234
235
236
237
            **hpt_params
        )
    else:
        tuner = BayesianOptimization(
            directory="BayOpt_{}_{}".format(loss, str(date.today())),
            max_trials=hypertun_trials,
            **hpt_params
        )
238
239
240

    print(tuner.search_space_summary())

241
242
243
244
245
246
247
248
249
250
251
    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]

252
    tuner.search(
253
254
        Xs,
        ys,
255
        epochs=n_epochs,
256
        validation_data=(Xvals, yvals),
257
258
        verbose=1,
        batch_size=256,
lucas_miranda's avatar
lucas_miranda committed
259
        callbacks=callbacks,
260
261
262
263
264
    )

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

lucas_miranda's avatar
lucas_miranda committed
265
266
    print(tuner.results_summary())

267
    return best_hparams, best_run