-
Notifications
You must be signed in to change notification settings - Fork 101
/
util.py
351 lines (291 loc) · 11 KB
/
util.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
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
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
345
346
347
348
349
350
351
# Copyright (c) Facebook, Inc. and its affiliates.
# This source code is licensed under the MIT license found in the
# LICENSE file in the root directory of this source tree.
import argparse
import os
import pathlib
import random
import subprocess
import sys
from collections import defaultdict
from typing import Any, Iterable, List, Optional
import numpy as np
import torch
from .distributed import maybe_init_distributed
common_opts = None
optimizer = None
summary_writer = None
def _populate_cl_params(arg_parser: argparse.ArgumentParser) -> argparse.ArgumentParser:
arg_parser.add_argument(
"--random_seed", type=int, default=None, help="Set random seed"
)
# trainer params
arg_parser.add_argument(
"--checkpoint_dir",
type=str,
default=None,
help="Where the checkpoints are stored",
)
arg_parser.add_argument(
"--preemptable",
default=False,
action="store_true",
help="If the flag is set, Trainer would always try to initialise itself from a checkpoint",
)
arg_parser.add_argument(
"--checkpoint_freq",
type=int,
default=0,
help="How often the checkpoints are saved",
)
arg_parser.add_argument(
"--validation_freq",
type=int,
default=1,
help="The validation would be run every `validation_freq` epochs",
)
arg_parser.add_argument(
"--n_epochs",
type=int,
default=10,
help="Number of epochs to train (default: 10)",
)
arg_parser.add_argument(
"--load_from_checkpoint",
type=str,
default=None,
help="If the parameter is set, model, trainer, and optimizer states are loaded from the "
"checkpoint (default: None)",
)
# cuda setup
arg_parser.add_argument(
"--no_cuda", default=False, help="disable cuda", action="store_true"
)
# dataset
arg_parser.add_argument(
"--batch_size",
type=int,
default=32,
help="Input batch size for training (default: 32)",
)
# optimizer
arg_parser.add_argument(
"--optimizer",
type=str,
default="adam",
help="Optimizer to use [adam, sgd, adagrad] (default: adam)",
)
arg_parser.add_argument(
"--lr", type=float, default=1e-2, help="Learning rate (default: 1e-2)"
)
arg_parser.add_argument(
"--update_freq",
type=int,
default=1,
help="Learnable weights are updated every update_freq batches (default: 1)",
)
# Channel parameters
arg_parser.add_argument(
"--vocab_size",
type=int,
default=10,
help="Number of symbols (terms) in the vocabulary including the end-of-sequence symbol <eos> (default: 10)",
)
arg_parser.add_argument(
"--max_len", type=int, default=1, help="Max length of the sequence (default: 1)"
)
# Setting up tensorboard
arg_parser.add_argument(
"--tensorboard", default=False, help="enable tensorboard", action="store_true"
)
arg_parser.add_argument(
"--tensorboard_dir", type=str, default="runs/", help="Path for tensorboard log"
)
arg_parser.add_argument(
"--distributed_port",
default=18363,
type=int,
help="Port to use in distributed learning",
)
arg_parser.add_argument(
"--fp16",
default=False,
help="Use mixed-precision for training/evaluating models",
action="store_true",
)
return arg_parser
def _get_params(
arg_parser: argparse.ArgumentParser, params: List[str]
) -> argparse.Namespace:
args = arg_parser.parse_args(params)
args.cuda = not args.no_cuda and torch.cuda.is_available()
# just to avoid confusion and be consistent
args.no_cuda = not args.cuda
args.device = torch.device("cuda" if args.cuda else "cpu")
args.distributed_context = maybe_init_distributed(args)
if args.fp16 and torch.__version__ < "1.6.0":
print("--fp16 is only supported with pytorch >= 1.6.0, please update!")
args.fp16 = False
return args
def init(
arg_parser: Optional[argparse.ArgumentParser] = None,
params: Optional[List[str]] = None,
) -> argparse.Namespace:
"""
Should be called before any code using egg; initializes the common components, such as
seeding logic etc.
:param arg_parser: An instance of argparse.ArgumentParser that is pre-populated if game-specific arguments.
`init` would add the commonly used arguments and parse the CL parameters. This allows us to easily obtain
commonly used parameters and have a full list of parameters obtained by a `--help` argument.
:param params: An optional list of parameters to be parsed against pre-defined frequently used parameters.
If set to None (default), command line parameters from sys.argv[1:] are used; setting to an empty list forces
to use default parameters.
"""
global common_opts
global optimizer
global summary_writer
if arg_parser is None:
arg_parser = argparse.ArgumentParser()
arg_parser = _populate_cl_params(arg_parser)
if params is None:
params = sys.argv[1:]
common_opts = _get_params(arg_parser, params)
if common_opts.random_seed is None:
common_opts.random_seed = random.randint(0, 2 ** 31)
elif common_opts.distributed_context:
common_opts.random_seed += common_opts.distributed_context.rank
_set_seed(common_opts.random_seed)
optimizers = {
"adam": torch.optim.Adam,
"sgd": torch.optim.SGD,
"adagrad": torch.optim.Adagrad,
}
if common_opts.optimizer in optimizers:
optimizer = optimizers[common_opts.optimizer]
else:
raise NotImplementedError(f"Unknown optimizer name {common_opts.optimizer}!")
if summary_writer is None and common_opts.tensorboard:
try:
from torch.utils.tensorboard import SummaryWriter
summary_writer = SummaryWriter(log_dir=common_opts.tensorboard_dir)
except ModuleNotFoundError:
raise ModuleNotFoundError(
"Cannot load tensorboard module; makes sure you installed everything required"
)
if common_opts.update_freq <= 0:
raise RuntimeError("update_freq should be an integer, >= 1.")
return common_opts
def close() -> None:
"""
Should be called at the end of the program - however, not required unless Tensorboard is used
"""
global summary_writer
if summary_writer:
summary_writer.close()
def get_opts() -> argparse.Namespace:
"""
:return: command line options
"""
global common_opts
return common_opts
def build_optimizer(params: Iterable) -> torch.optim.Optimizer:
return optimizer(params, lr=get_opts().lr)
def get_summary_writer() -> "torch.utils.SummaryWriter":
"""
:return: Returns an initialized instance of torch.util.SummaryWriter
"""
global summary_writer
return summary_writer
def _set_seed(seed) -> None:
"""
Seeds the RNG in python.random, torch {cpu/cuda}, numpy.
:param seed: Random seed to be used
>>> _set_seed(10)
>>> random.randint(0, 100), torch.randint(0, 100, (1,)).item(), np.random.randint(0, 100)
(73, 37, 9)
>>> _set_seed(10)
>>> random.randint(0, 100), torch.randint(0, 100, (1,)).item(), np.random.randint(0, 100)
(73, 37, 9)
"""
random.seed(seed)
torch.manual_seed(seed)
np.random.seed(seed)
if torch.cuda.is_available():
torch.cuda.manual_seed_all(seed)
def move_to(x: Any, device: torch.device) -> Any:
"""
Simple utility function that moves a tensor or a dict/list/tuple of (dict/list/tuples of ...) tensors
to a specified device, recursively.
:param x: tensor, list, tuple, or dict with values that are lists, tuples or dicts with values of ...
:param device: device to be moved to
:return: Same as input, but with all tensors placed on device. Non-tensors are not affected.
For dicts, the changes are done in-place!
"""
if hasattr(x, "to"):
return x.to(device)
if isinstance(x, list) or isinstance(x, tuple):
return [move_to(i, device) for i in x]
if isinstance(x, dict) or isinstance(x, defaultdict):
for k, v in x.items():
x[k] = move_to(v, device)
return x
return x
def load_interactions(file_path: str):
file_path = pathlib.Path(file_path)
assert (
file_path.exists()
), f"{file_path} does not exist. Interactions cannot be loaded"
try:
return torch.load(file_path)
except FileNotFoundError:
print(f"{file_path} was an invalid path to load interactions.")
exit(1)
def find_lengths(messages: torch.Tensor) -> torch.Tensor:
"""
:param messages: A tensor of term ids, encoded as Long values, of size (batch size, max sequence length).
:returns A tensor with lengths of the sequences, including the end-of-sequence symbol <eos> (in EGG, it is 0).
If no <eos> is found, the full length is returned (i.e. messages.size(1)).
>>> messages = torch.tensor([[1, 1, 0, 0, 0, 1], [1, 1, 1, 10, 100500, 5]])
>>> lengths = find_lengths(messages)
>>> lengths
tensor([3, 6])
"""
max_k = messages.size(1)
zero_mask = messages == 0
# a bit involved logic, but it seems to be faster for large batches than slicing batch dimension and
# querying torch.nonzero()
# zero_mask contains ones on positions where 0 occur in the outputs, and 1 otherwise
# zero_mask.cumsum(dim=1) would contain non-zeros on all positions after 0 occurred
# zero_mask.cumsum(dim=1) > 0 would contain ones on all positions after 0 occurred
# (zero_mask.cumsum(dim=1) > 0).sum(dim=1) equates to the number of steps happened after 0 occured (including it)
# max_k - (zero_mask.cumsum(dim=1) > 0).sum(dim=1) is the number of steps before 0 took place
lengths = max_k - (zero_mask.cumsum(dim=1) > 0).sum(dim=1)
lengths.add_(1).clamp_(max=max_k)
return lengths
def setup_print_for_distributed(is_master: bool):
"""Prevents non-master processes from printing to stdout unless explicitly requested."""
import builtins as __builtin__
builtin_print = __builtin__.print
def print(*args, **kwargs):
force = kwargs.pop("force", False)
if is_master or force:
builtin_print(*args, **kwargs)
__builtin__.print = print
def get_sha():
"""Computes and returns info about current state of working directory and staging area."""
cwd = os.path.dirname(os.path.abspath(__file__))
def _run(command):
return subprocess.check_output(command, cwd=cwd).decode("ascii").strip()
sha = "N/A"
diff = "clean"
branch = "N/A"
try:
sha = _run(["git", "rev-parse", "HEAD"])
subprocess.check_output(["git", "diff"], cwd=cwd)
diff = _run(["git", "diff-index", "HEAD"])
diff = "has uncommited changes" if diff else "clean"
branch = _run(["git", "rev-parse", "--abbrev-ref", "HEAD"])
except Exception:
pass
message = f"sha: {sha}, status: {diff}, branch: {branch}"
return message