forked from google-deepmind/tapnet
-
Notifications
You must be signed in to change notification settings - Fork 0
/
experiment_utils.py
246 lines (207 loc) · 8.61 KB
/
experiment_utils.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
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
184
185
186
187
188
189
190
191
192
193
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
233
234
235
236
237
238
239
240
241
242
243
244
245
246
# Copyright 2023 DeepMind Technologies Limited
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Logging and other experiment utilities."""
import os
from typing import Mapping, Optional
import jax
from jaxline import utils
from ml_collections import config_dict
import numpy as np
import optax
import tensorflow as tf
from tapnet import optimizers
def get_lr_schedule(
total_steps: int,
optimizer_config: config_dict.ConfigDict,
) -> optax.Schedule:
"""Build the LR schedule function."""
base_lr = optimizer_config.base_lr
schedule_type = optimizer_config.schedule_type
if schedule_type == 'cosine':
warmup_steps = (optimizer_config.cosine_decay_kwargs.warmup_steps)
# Batch scale the other lr values as well:
init_value = optimizer_config.cosine_decay_kwargs.init_value
end_value = optimizer_config.cosine_decay_kwargs.end_value
schedule_fn = optax.warmup_cosine_decay_schedule(
init_value=init_value,
peak_value=base_lr,
warmup_steps=warmup_steps,
decay_steps=total_steps,
end_value=end_value)
elif schedule_type == 'constant_cosine':
# Convert end_value to alpha, used by cosine_decay_schedule.
alpha = optimizer_config.constant_cosine_decay_kwargs.end_value / base_lr
# Number of steps spent in constant phase.
constant_steps = int(
optimizer_config.constant_cosine_decay_kwargs.constant_fraction *
total_steps)
decay_steps = total_steps - constant_steps
constant_phase = optax.constant_schedule(value=base_lr)
decay_phase = optax.cosine_decay_schedule(
init_value=base_lr, decay_steps=decay_steps, alpha=alpha)
schedule_fn = optax.join_schedules(
schedules=[constant_phase, decay_phase], boundaries=[constant_steps])
else:
raise ValueError(f'Unknown learning rate schedule: {schedule_type}')
return schedule_fn
def make_optimizer(
optimizer_config: config_dict.ConfigDict,
lr_schedule: optax.Schedule,
) -> optax.GradientTransformation:
"""Construct the optax optimizer with given LR schedule."""
# Decay learned position embeddings by default.
weight_decay_exclude_names = ['b']
optax_chain = []
if optimizer_config.max_norm > 0:
optax_chain.append(optax.clip_by_global_norm(optimizer_config.max_norm))
if optimizer_config.optimizer == 'sgd':
optax_chain.extend([
optax.trace(**optimizer_config.sgd_kwargs),
optimizers.add_weight_decay(
optimizer_config.weight_decay,
exclude_names=weight_decay_exclude_names)
])
elif optimizer_config.optimizer == 'adam':
optax_chain.extend([
optax.scale_by_adam(**optimizer_config.adam_kwargs),
optimizers.add_weight_decay(
optimizer_config.weight_decay,
exclude_names=weight_decay_exclude_names)
])
else:
raise ValueError(f'Undefined optimizer {optimizer_config.optimizer}')
optax_chain.extend([
optax.scale_by_schedule(lr_schedule),
optax.scale(-1),
])
optimizer = optax.chain(*optax_chain)
optimizer = optax.apply_if_finite(optimizer, max_consecutive_errors=5)
return optimizer
class NumpyFileCheckpointer(utils.Checkpointer):
"""A Jaxline checkpointer which saves to numpy files on disk."""
def __init__(self, config: config_dict.ConfigDict, mode: str):
self._checkpoint_file = os.path.join(config.checkpoint_dir,
'checkpoint.npy')
self._checkpoint_state = config_dict.ConfigDict()
del mode
def get_experiment_state(self, ckpt_series: str) -> config_dict.ConfigDict:
"""Returns the experiment state for a given checkpoint series."""
if ckpt_series != 'latest':
raise ValueError('multiple checkpoint series are not supported')
return self._checkpoint_state
def save(self, ckpt_series: str) -> None:
"""Saves the checkpoint."""
if ckpt_series != 'latest':
raise ValueError('multiple checkpoint series are not supported')
exp_mod = self._checkpoint_state.experiment_module
global_step = self._checkpoint_state.global_step
f_np = lambda x: np.array(jax.device_get(utils.get_first(x)))
to_save = {}
for attr, name in exp_mod.CHECKPOINT_ATTRS.items():
if name == 'global_step':
raise ValueError(
'global_step attribute would overwrite jaxline global step')
np_params = jax.tree_map(f_np, getattr(exp_mod, attr))
to_save[name] = np_params
to_save['global_step'] = global_step
with tf.io.gfile.GFile(self._checkpoint_file + '_tmp', 'wb') as fp:
np.save(fp, to_save)
tf.io.gfile.rename(
self._checkpoint_file + '_tmp',
self._checkpoint_file,
overwrite=True,
)
def can_be_restored(self, ckpt_series: str) -> bool:
"""Returns whether or not a given checkpoint series can be restored."""
if ckpt_series != 'latest':
raise ValueError('multiple checkpoint series are not supported')
return tf.io.gfile.exists(self._checkpoint_file)
def restore(self, ckpt_series: str) -> None:
"""Restores the checkpoint."""
experiment_state = self.get_experiment_state(ckpt_series)
with tf.io.gfile.GFile(self._checkpoint_file, 'rb') as fp:
ckpt_state = np.load(fp, allow_pickle=True).item()
experiment_state.global_step = int(ckpt_state['global_step'])
exp_mod = experiment_state.experiment_module
for attr, name in exp_mod.CHECKPOINT_ATTRS.items():
setattr(exp_mod, attr, utils.bcast_local_devices(ckpt_state[name]))
def restore_path(self, ckpt_series: str) -> Optional[str]:
"""Returns the restore path for the checkpoint, or None."""
if not self.can_be_restored(ckpt_series):
return None
return self._checkpoint_file
def wait_for_checkpointing_to_finish(self) -> None:
"""Waits for any async checkpointing to complete."""
@classmethod
def create(
cls,
config: config_dict.ConfigDict,
mode: str,
) -> utils.Checkpointer:
return cls(config, mode)
def default_color_augmentation_fn(
inputs: Mapping[str, tf.Tensor]) -> Mapping[str, tf.Tensor]:
"""Standard color augmentation for videos.
Args:
inputs: A DatasetElement containing the item 'video' which will have
augmentations applied to it.
Returns:
A DatasetElement with all the same data as the original, except that
the video has augmentations applied.
"""
zero_centering_image = True
prob_color_augment = 0.8
prob_color_drop = 0.2
frames = inputs['video']
if frames.dtype != tf.float32:
raise ValueError('`frames` should be in float32.')
def color_augment(video: tf.Tensor) -> tf.Tensor:
"""Do standard color augmentations."""
# Note the same augmentation will be applied to all frames of the video.
if zero_centering_image:
video = 0.5 * (video + 1.0)
video = tf.image.random_brightness(video, max_delta=32. / 255.)
video = tf.image.random_saturation(video, lower=0.6, upper=1.4)
video = tf.image.random_contrast(video, lower=0.6, upper=1.4)
video = tf.image.random_hue(video, max_delta=0.2)
video = tf.clip_by_value(video, 0.0, 1.0)
if zero_centering_image:
video = 2 * (video-0.5)
return video
def color_drop(video: tf.Tensor) -> tf.Tensor:
video = tf.image.rgb_to_grayscale(video)
video = tf.tile(video, [1, 1, 1, 3])
return video
# Eventually applies color augmentation.
coin_toss_color_augment = tf.random.uniform(
[], minval=0, maxval=1, dtype=tf.float32)
frames = tf.cond(
pred=tf.less(coin_toss_color_augment,
tf.cast(prob_color_augment, tf.float32)),
true_fn=lambda: color_augment(frames),
false_fn=lambda: frames)
# Eventually applies color drop.
coin_toss_color_drop = tf.random.uniform(
[], minval=0, maxval=1, dtype=tf.float32)
frames = tf.cond(
pred=tf.less(coin_toss_color_drop, tf.cast(prob_color_drop, tf.float32)),
true_fn=lambda: color_drop(frames),
false_fn=lambda: frames)
result = {**inputs}
result['video'] = frames
return result
def add_default_data_augmentation(ds: tf.data.Dataset) -> tf.data.Dataset:
return ds.map(
default_color_augmentation_fn, num_parallel_calls=tf.data.AUTOTUNE)