def logging_loop(self, it):
""" Logging and checkpointing stuff """
if it % self.args.validate_every == 0:
self.mapper.eval()
self.navigator.eval()
loss_val_seen = self.validate("val_seen")
loss_val_unseen = self.validate("val_unseen")
self.prev_time, time_taken = utils.time_it(self.prev_time)
print("Iteration: %d Loss: %f Val Seen Loss: %f Val Unseen Loss: %f Time: %0.2f secs"
%(it, self.loss.item(), loss_val_seen.item(), loss_val_unseen.item(), time_taken))
if self.visdom:
# visdom: X, Y, key, line_name, x_label, y_label, fig_title
self.visdom.line(it, self.loss.item(), "train_loss", "Train Loss", "Iterations", "Loss", title=" Train Phase")
self.visdom.line(it, loss_val_seen.item(), "val_loss", "Val Seen Loss", "Iterations", "Loss", title="Val Phase")
self.visdom.line(it, loss_val_unseen.item(), "val_loss", "Val Unseen Loss", "Iterations", "Loss", title="Val Phase")
self.mapper.train()
self.navigator.train()
elif it % self.args.log_every == 0:
self.prev_time, time_taken = utils.time_it(self.prev_time)
print("Iteration: %d Loss: %f Time: %0.2f secs" % (it, self.loss.item(), time_taken))
if self.visdom:
self.visdom.line(it, self.loss.item(), "train_loss", "Train Loss", "Iterations", "Loss", title="Train Phase")
if it % self.args.checkpoint_every == 0:
saver = {"mapper": self.mapper.state_dict(),
"navigator": self.navigator.state_dict(),
"args": self.args}
dir = "%s/%s" % (self.args.snapshot_dir, self.args.exp_name)
if not os.path.exists(dir):
os.makedirs(dir)
torch.save(
saver, "%s/%s_%d" % (dir, self.args.exp_name, it)
)
if self.visdom:
self.visdom.save()
def __init__(self, args):
super(PanoSimulator, self).__init__()
self.args = args
if self.args.preloading:
print("Simulator images will be preloaded")
prev_time = time.time()
self.setPreloadingEnabled(self.args.preloading)
self.setDepthEnabled(True)
self.setRestrictedNavigation(False)
self.setBatchSize(self.args.batch_size)
self.setCacheSize(self.args.cache_size)
self.setCameraResolution(self.args.image_width, self.args.image_height)
self.setCameraVFOV(self.args.vfov)
self.initialize()
self._reset_visited_goal()
prev_time, time_taken = utils.time_it(prev_time)
print("Simulator initialized: %0.2f secs" % time_taken)
# 1. Combine multi line messages
# 4. Map phone numbers to users
# 2. Extract emojis
# 5. Sentiment analysis
# FILES AND LOCATIONS # of images, videos, voicemessages and locations
# GIF omitted
# image omitted
# video omitted
# audio omitted
# Contact card omitted
# Location: https://maps.google.com/?q=-13.523442,-71.950256
# TIMESPAN: how long did the chat go
# TIMELINE: when and how much you where messaging
# TOTAL NUMBERS days you are chatting, message/word/letter count
# TOPS most active day
if __name__ == "__main__":
with time_it('Reading'):
with open('_chat.txt', 'r') as f:
chat = f.readlines()
with open('numbers.json', 'r') as f:
user_number_map = json.load(f)
with time_it('Parsing'):
grouped_messages = Message.parse_chat(chat, user_number_map)
with time_it('Stats'):
stats = compute_stats(grouped_messages)
pprint(stats)
shuffle=(train_sampler is None),
num_workers=n_workers,
pin_memory=True,
sampler=train_sampler)
val_loader = torch.utils.data.DataLoader(datasets.ImageFolder(
valdir,
transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
normalize,
])),
batch_size=batch_size,
shuffle=False,
num_workers=n_workers,
pin_memory=True)
return train_loader, val_loader
if __name__ == "__main__":
# a, b = get_future_bar_classification_data(16)
# test_dataset_for_loop()
from utils import time_it
time_it(test_dataset_for_loop
# test_dataset_loader
# get_cifar_10, 10, shuffle=True
)
def logging_loop(self, it):
""" Logging and checkpointing stuff """
if it % self.args.validate_every == 0:
self.mapper.eval()
self.model.eval()
self.policy.eval()
scores_train = self.validate("train")
scores_val_seen = self.validate("val_seen")
scores_val_unseen = self.validate("val_unseen")
self.prev_time, time_taken = utils.time_it(self.prev_time)
print(
"Iteration: %d Loss: %f Train Success: %f Val Seen Success: %f Val Unseen Success: %f Time: %0.2f secs"
% (it, self.loss.item(), scores_train['success'],
scores_val_seen['success'], scores_val_unseen['success'],
time_taken))
if self.visdom:
# visdom: X, Y, key, line_name, x_label, y_label, fig_title
self.visdom.line(it,
self.loss.item(),
"train_loss",
"Train Loss",
"Iterations",
"Loss",
title=" Train Phase")
units = {
'length': 'm',
'error': 'm',
'oracle success': '%',
'success': '%',
'spl': '%'
}
sub = self.args.validation_iterations * self.args.batch_size
for metric, score in scores_train.items():
m = metric.title()
self.visdom.line(it,
score,
metric,
"Train (%d)" % sub,
"Iterations",
units[metric],
title=m)
for metric, score in scores_val_seen.items():
m = metric.title()
self.visdom.line(it,
score,
metric,
"Val Seen (%d)" % sub,
"Iterations",
units[metric],
title=m)
for metric, score in scores_val_unseen.items():
m = metric.title()
self.visdom.line(it,
score,
metric,
"Val Unseen (%d)" % sub,
"Iterations",
units[metric],
title=m)
self.mapper.train()
self.model.train()
self.policy.train()
elif it % self.args.log_every == 0:
self.prev_time, time_taken = utils.time_it(self.prev_time)
print("Iteration: %d Loss: %f Time: %0.2f secs" %
(it, self.loss.item(), time_taken))
if self.visdom:
self.visdom.line(it,
self.loss.item(),
"train_loss",
"Train Loss",
"Iterations",
"Loss",
title="Train Phase")
if it % self.args.checkpoint_every == 0:
saver = {
"mapper": self.mapper.state_dict(),
"args": self.args,
"filter": self.model.state_dict(),
"policy": self.policy.state_dict()
}
dir = "%s/%s" % (self.args.snapshot_dir, self.args.exp_name)
if not os.path.exists(dir):
os.makedirs(dir)
torch.save(saver, "%s/%s_%d" % (dir, self.args.exp_name, it))
if self.visdom:
self.visdom.save()