from common.argparser import argparser
from common.arguments import Arguments
from common.utils import get_paths
from common.approximation import get_dga_sdirs
import pickle as pkl
args = Arguments(argparser())
paths = get_paths(args)
print('Loading data: {}'.format(paths.data_path))
X_trains, _, y_trains, _, meta = pkl.load(open(paths.data_path, 'rb'))
sdirs = get_dga_sdirs(args, X_trains, y_trains)
print('Saving:', paths.dga_path)
if not args.dry_run:
pkl.dump(sdirs, open(paths.dga_path, 'wb'))
from data.loader import get_loader
from models.train import distributed_train, test
from models.utils import get_model
from viz.training_plots import training_plots
print = functools.partial(print, flush=True)
torch.set_printoptions(linewidth=120)
# ------------------------------------------------------------------------------
# Setups
# ------------------------------------------------------------------------------
args = Arguments(argparser())
hook = sy.TorchHook(torch)
device = get_device(args)
paths = get_paths(args, distributed=True)
log_file, std_out = init_logger(paths.log_file, args.dry_run, args.load_model)
if os.path.exists(paths.tb_path):
shutil.rmtree(paths.tb_path)
tb = SummaryWriter(paths.tb_path)
print('+' * 80)
print(paths.model_name)
print('+' * 80)
print(args.__dict__)
print('+' * 80)
# prepare graph and data
_, workers = get_fl_graph(hook, args.num_workers)
print('Loading data: {}'.format(paths.data_path))