def eval_models(args: Argument):
args.mode = 'test'
dc, lc, tc, model_dir = get_config_list(args)
modes = ['test']
dataloader = {
'test':
get_trajectory_data_loader(dc,
test=True,
batch_size=args.bsize,
num_workers=args.num_workers,
shuffle=True)
}
run_every = {'test': 1}
gn_wrapper = fetch_model_iterator(lc, args)
trainer = train.Trainer(gn_wrapper, modes, dataloader, run_every, tc)
output = trainer.eval(dataloader['test'])
return trainer.num_iter, output
def train_model(args: Argument):
args.mode = 'train'
dc, lc, tc, _ = get_config_list(args)
gn_wrapper = fetch_model_iterator(lc, args)
modes = ['train', 'test']
dataloader = {
m: get_trajectory_data_loader(dc,
test=m == 'train',
batch_size=args.bsize,
num_workers=args.num_workers,
shuffle=True)
for m in modes
}
run_every = {'train': 1, 'test': args.test_every}
trainer = train.Trainer(gn_wrapper, modes, dataloader, run_every, tc)
train_winding = False
train_trajectory = True
trainer.train(train_winding, train_trajectory)
trainer.save(train_winding, train_trajectory)