Spaces:
Sleeping
Sleeping
File size: 6,079 Bytes
c53ddec |
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 |
# import imageio
import logging
import os
from collections import defaultdict
from timeit import default_timer
import torch
from torch.nn import functional as F
from tqdm import trange
from disvae.utils.modelIO import save_model
TRAIN_LOSSES_LOGFILE = "train_losses.log"
class Trainer:
"""
Class to handle training of model.
Parameters
----------
model: disvae.vae.VAE
optimizer: torch.optim.Optimizer
loss_f: disvae.models.BaseLoss
Loss function.
device: torch.device, optional
Device on which to run the code.
logger: logging.Logger, optional
Logger.
save_dir : str, optional
Directory for saving logs.
gif_visualizer : viz.Visualizer, optional
Gif Visualizer that should return samples at every epochs.
is_progress_bar: bool, optional
Whether to use a progress bar for training.
"""
def __init__(
self,
model,
optimizer,
loss_f,
device=torch.device("cpu"),
logger=logging.getLogger(__name__),
save_dir="results",
gif_visualizer=None,
is_progress_bar=True,
):
self.device = device
self.model = model.to(self.device)
self.loss_f = loss_f
self.optimizer = optimizer
self.save_dir = save_dir
self.is_progress_bar = is_progress_bar
self.logger = logger
self.losses_logger = LossesLogger(
os.path.join(self.save_dir, TRAIN_LOSSES_LOGFILE)
)
self.gif_visualizer = gif_visualizer
self.logger.info("Training Device: {}".format(self.device))
def __call__(self, data_loader, epochs=10, checkpoint_every=10):
"""
Trains the model.
Parameters
----------
data_loader: torch.utils.data.DataLoader
epochs: int, optional
Number of epochs to train the model for.
checkpoint_every: int, optional
Save a checkpoint of the trained model every n epoch.
"""
start = default_timer()
self.model.train()
for epoch in range(epochs):
storer = defaultdict(list)
mean_epoch_loss = self._train_epoch(data_loader, storer, epoch)
self.logger.info(
"Epoch: {} Average loss per image: {:.2f}".format(
epoch + 1, mean_epoch_loss
)
)
self.losses_logger.log(epoch, storer)
if self.gif_visualizer is not None:
self.gif_visualizer()
if epoch % checkpoint_every == 0:
save_model(
self.model, self.save_dir, filename="model-{}.pt".format(epoch)
)
if self.gif_visualizer is not None:
self.gif_visualizer.save_reset()
self.model.eval()
delta_time = (default_timer() - start) / 60
self.logger.info("Finished training after {:.1f} min.".format(delta_time))
def _train_epoch(self, data_loader, storer, epoch):
"""
Trains the model for one epoch.
Parameters
----------
data_loader: torch.utils.data.DataLoader
storer: dict
Dictionary in which to store important variables for vizualisation.
epoch: int
Epoch number
Return
------
mean_epoch_loss: float
Mean loss per image
"""
epoch_loss = 0.0
kwargs = dict(
desc="Epoch {}".format(epoch + 1),
leave=False,
disable=not self.is_progress_bar,
)
with trange(len(data_loader), **kwargs) as t:
for _, (data, _) in enumerate(data_loader):
iter_loss = self._train_iteration(data, storer)
epoch_loss += iter_loss
t.set_postfix(loss=iter_loss)
t.update()
mean_epoch_loss = epoch_loss / len(data_loader)
return mean_epoch_loss
def _train_iteration(self, data, storer):
"""
Trains the model for one iteration on a batch of data.
Parameters
----------
data: torch.Tensor
A batch of data. Shape : (batch_size, channel, height, width).
storer: dict
Dictionary in which to store important variables for vizualisation.
"""
batch_size, channel, height, width = data.size()
data = data.to(self.device)
try:
recon_batch, latent_dist, latent_sample = self.model(data)
loss = self.loss_f(
data,
recon_batch,
latent_dist,
self.model.training,
storer,
latent_sample=latent_sample,
)
self.optimizer.zero_grad()
loss.backward()
self.optimizer.step()
except ValueError:
# for losses that use multiple optimizers (e.g. Factor)
loss = self.loss_f.call_optimize(data, self.model, self.optimizer, storer)
return loss.item()
class LossesLogger(object):
"""Class definition for objects to write data to log files in a
form which is then easy to be plotted.
"""
def __init__(self, file_path_name):
"""Create a logger to store information for plotting."""
if os.path.isfile(file_path_name):
os.remove(file_path_name)
self.logger = logging.getLogger("losses_logger")
self.logger.setLevel(1) # always store
file_handler = logging.FileHandler(file_path_name)
file_handler.setLevel(1)
self.logger.addHandler(file_handler)
header = ",".join(["Epoch", "Loss", "Value"])
self.logger.debug(header)
def log(self, epoch, losses_storer):
"""Write to the log file"""
for k, v in losses_storer.items():
log_string = ",".join(str(item) for item in [epoch, k, mean(v)])
self.logger.debug(log_string)
# HELPERS
def mean(l):
"""Compute the mean of a list"""
return sum(l) / len(l)
|