def do_train_epoch(i: int) -> Tuple[float, str]:
train_epoch(ref_model, optimizer, train_loader, i, use_cuda,
loss_func=lossfunc, log_interval=100)
time_stamp = get_timestamp()
avg_loss, _ = eval_epoch(ref_model, val_loader, i, use_cuda,
loss_func=lossfunc)
return avg_loss, time_stamp
def do_train_epoch(i: int) -> Tuple[float, str]:
train_epoch(msg_model, optimizer, train_loader, i, use_cuda,
loss_func=lossfunc,
output_ops=lambda x: x[0].squeeze(1), log_interval=100)
time_stamp = get_timestamp()
avg_loss, _ = eval_epoch(msg_model, val_loader, i, use_cuda,
loss_func=lossfunc,
input_ops=lambda x: x.unsqueeze(-1),
output_ops=lambda x: x[0].squeeze(1))
return avg_loss, time_stamp
mkdirp(data_dir)
mkdirp(save_dir)
mkdirp(log_dir)
# prepare proper loss function
lossfunc = nnf.nll_loss
# start logger
log_file = '{0}_{1}_{2}_{3}.log'.format(
proto_name,
'-'.join([str(i) for i in (frequency, train_batch, val_batch,
test_batch)]),
'-'.join([infer_method, dataset_flavor]),
get_timestamp())
log_title = log_file[:-4]
logger = Log(log_dir + log_file)
logger.start(log_title)
logger.start_intercept()
# check cuda availablility when needed
if use_cuda:
check_cuda()
else:
print('Currently using cpu device')
# set up dataset
mkdirp(data_dir)
mkdirp(save_dir)
mkdirp(log_dir)
# prepare proper loss function
lossfunc = nnf.nll_loss
# start logger
log_file = '{0}_{1}_{2}_{3}.log'.format(
"LocalMininet" if hidden_layers is None else '-'.join(
[str(i) for i in hidden_layers]),
'-'.join([str(i) for i in (train_batch, val_batch, test_batch,
activ_func.__name__)]),
dataset_flavor, get_timestamp())
log_title = log_file[:-4]
logger = Log(log_dir + log_file)
logger.start(log_title)
logger.start_intercept()
# check cuda availablility when needed
if use_cuda:
check_cuda()
# set up dataset
if dataset_flavor == 'MNIST':
((train_loader, val_loader, test_loader),
(nb_train, nb_val, nb_test)) = get_MNIST_dataloaders(