Spaces:
Sleeping
Sleeping
File size: 10,371 Bytes
957e2dc | 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 | import torch
import torch.nn as nn
from copy import deepcopy
import contextlib
import math
import time
import logging
import sys
import json
from typing import Union, Dict, Any
from pathlib import Path
from torch.utils.tensorboard import SummaryWriter
from src.constants import RUNS_DIR
from src.utils.filesystem import ensure_dir
from src.utils.plotting import *
################################################################################
# Logging utility with optional TensorBoard support
################################################################################
class Writer:
"""
Handles file, console, and TensorBoard logging
"""
def __init__(self,
root_dir: Union[str, Path] = RUNS_DIR,
name: str = None,
use_tb: bool = False,
log_iter: int = 100,
use_timestamp: bool = True,
log_images: bool = False,
**kwargs
):
"""
Configure logging.
:param root_dir: root logging directory
:param name: descriptive name for run
:param use_tb: if True, use TensorBoard
:param log_iter: iterations between logging
"""
# generate run-specific name and create directory
run_name = f'{name}'
if use_timestamp:
run_name += f'_{time.strftime("%m-%d-%H_%M_%S")}'
self.run_dir = Path(root_dir) / run_name
ensure_dir(self.run_dir)
# prepare checkpoint directory
self.checkpoint_dir = self.run_dir / 'checkpoints'
# log to TensorBoard
self.use_tb = use_tb
self.log_iter = log_iter
self.writer = SummaryWriter(
log_dir=str(self.run_dir),
flush_secs=20,
) if self.use_tb else None
# log to console and file 'log.txt'
self.logger = logging.getLogger(run_name)
self.logger.setLevel(logging.INFO)
self.logger.addHandler(
logging.StreamHandler(sys.stdout)
)
self.logger.addHandler(
logging.FileHandler(self.run_dir / 'log.txt')
)
# to avoid segmentation faults, it may be necessary to skip image
# logging
self.log_images = log_images
# self.logger.info(f'Logging to {self.run_dir}')
# disable Matplotlib logging
logging.getLogger('matplotlib.font_manager').disabled = True
def get_run_dir(self):
return str(self.run_dir)
def log_info(self, info: str):
"""
Log generic statements
"""
self.logger.info(info)
def _dict_to_str(self, d: dict):
"""Recursively cast dictionary entries to strings"""
d_out = {}
for k, v in d.items():
if isinstance(v, dict):
d_out[k] = self._dict_to_str(v)
elif not isinstance(v, (float, int, bool)):
d_out[k] = str(v)
else:
d_out[k] = v
return d_out
def log_config(self,
config: Union[dict, str],
tag: str = "config",
path: Union[str, Path] = None):
"""Save config file for run, given dictionary"""
path = path if path is not None else self.run_dir / f'{tag}.conf'
with open(path, "w") as out_config:
self.logger.info(f'Saving config to {path}')
if isinstance(config, dict):
config = self._dict_to_str(config)
json.dump(config, out_config, indent=4)
else:
out_config.write(config)
def log_scalar(self, x: torch.Tensor, tag: str, global_step: int = 0):
"""
Log scalar
"""
# only log at specified iterations
if not self.log_iter or global_step % self.log_iter:
return
# log scalar to file and console
self.logger.info(f'iter {global_step}\t{tag}: {x}')
# if TensorBoard is enabled
if self.use_tb:
self.writer.add_scalar(f'{tag}', x, global_step=global_step)
self.writer.flush()
def log_logits(self,
x: torch.Tensor,
target: int = None,
tag: str = None,
global_step: int = 0):
"""
Log class scores (logits)
"""
# only log at specified iterations
if not self.log_iter or global_step % self.log_iter:
return
# log plot to TensorBoard
if self.use_tb and self.log_images:
self.writer.add_image(
f"{tag}", plot_logits(x, target),
global_step=global_step
)
self.writer.flush()
def log_audio(self,
x: torch.Tensor,
tag: str,
global_step: int = 0,
sample_rate: int = 16000,
scale: Union[int, float] = 1.0):
"""
Given a single audio waveform, log a normalized recording, waveform
plot, and spectrogram plot to TensorBoard
"""
# only log at specified iterations and if TensorBoard is enabled
if not self.log_iter or global_step % self.log_iter or not self.use_tb:
return
# normalize and log audio recording
normalized = (scale / torch.max(
torch.abs(x) + 1e-12, dim=-1, keepdim=True
)[0]) * x * 0.95
self.writer.add_audio(f"{tag}-audio",
normalized,
sample_rate=sample_rate,
global_step=global_step)
if self.log_images:
# log waveform
self.writer.add_image(f"{tag}-waveform",
plot_waveform(x, scale),
global_step=global_step)
# log spectrogram
self.writer.add_image(f"{tag}-spectrogram",
plot_spectrogram(x),
global_step=global_step)
# flush
self.writer.flush()
def log_norm(self,
x: torch.Tensor,
tag: str,
global_step: int = 0):
"""
Plot norm of input tensor
"""
# only log at specified iterations and if TensorBoard is enabled
if not self.log_iter or global_step % self.log_iter or not self.use_tb:
return
# log norms
norm_2 = torch.norm(x, p=2)
norm_inf = torch.norm(x, p=float('inf'))
self.writer.add_scalar(f'{tag}/norm-2', norm_2, global_step=global_step)
self.writer.add_scalar(f'{tag}/norm-inf', norm_inf, global_step=global_step)
self.writer.flush()
def log_image(self, image: torch.Tensor, tag: str, global_step: int = 0):
"""
Log image plot
"""
if not self.log_images:
return
# only log at specified iterations and if TensorBoard is enabled
if not self.log_iter or global_step % self.log_iter or not self.use_tb:
return
self.writer.add_image(
tag,
image,
global_step
)
self.writer.flush()
def log_filter(self,
amplitudes: torch.Tensor,
tag: str,
global_step: int = 0):
"""
Plot filter controls
"""
# only log at specified iterations and if TensorBoard is enabled
if not self.log_iter or global_step % self.log_iter or not self.use_tb:
return
if self.log_images:
self.writer.add_image(
f'filter_controls/{tag}',
plot_filter(amplitudes),
global_step
)
self.writer.flush()
@staticmethod
def bytes_to_gb(size_bytes: int):
"""
Code from: https://stackoverflow.com/a/14822210
"""
if size_bytes == 0:
return "0B"
size_name = ("B", "KB", "MB", "GB", "TB", "PB", "EB", "ZB", "YB")
i = int(math.floor(math.log(size_bytes, 1024)))
p = math.pow(1024, i)
s = round(size_bytes / p, 2)
return "%s %s" % (s, size_name[i])
def log_cuda_memory(self, device: int = 0):
total_memory = self.bytes_to_gb(
torch.cuda.get_device_properties(device).total_memory
)
reserved_memory = self.bytes_to_gb(
torch.cuda.memory_reserved(device)
)
allocated_memory = self.bytes_to_gb(
torch.cuda.memory_allocated(device)
)
free_memory = self.bytes_to_gb(
torch.cuda.memory_reserved(device) - torch.cuda.memory_allocated(
device
)
)
self.logger.info(f'\nMemory management:\n'
f'------------------\n'
f'Total: {total_memory}\n'
f'Reserved: {reserved_memory}\n'
f'Allocated: {allocated_memory}\n'
f'Free: {free_memory}\n')
def checkpoint(self,
checkpoint: Union[nn.Module, Dict[str, Any]],
tag: str,
global_step: int = None
):
"""
Given nn.Module object or state dictionary, save to disk
"""
ensure_dir(self.checkpoint_dir)
if global_step is not None:
filename = f'{tag}_{global_step}.pt'
else:
filename = f'{tag}.pt'
if isinstance(checkpoint, nn.Module):
checkpoint = checkpoint.state_dict()
torch.save(
checkpoint,
self.checkpoint_dir / filename
)
@contextlib.contextmanager
def force_logging(self):
"""Force Writer to log by temporarily overriding logging interval"""
log_iter = self.log_iter
self.log_iter = 1
yield
self.log_iter = log_iter
|