params['nll_loss_type'] = 'gaussian'
train_data = SmallSynthData(args.data_path, 'train', params)
val_data = SmallSynthData(args.data_path, 'val', params)
model = model_builder.build_model(params)
if args.mode == 'train':
with train_utils.build_writers(args.working_dir) as (train_writer,
val_writer):
train.train(model, train_data, val_data, params, train_writer,
val_writer)
elif args.mode == 'eval':
test_data = SmallSynthData(args.data_path, 'test', params)
forward_pred = 50 - args.test_burn_in_steps
test_mse = evaluate.eval_forward_prediction(model, test_data,
args.test_burn_in_steps,
forward_pred, params)
path = os.path.join(args.working_dir,
args.error_out_name % args.test_burn_in_steps)
np.save(path, test_mse.cpu().numpy())
test_mse_1 = test_mse[0].item()
test_mse_15 = test_mse[14].item()
test_mse_25 = test_mse[24].item()
print("FORWARD PRED RESULTS:")
print("\t1 STEP: ", test_mse_1)
print("\t15 STEP: ", test_mse_15)
print("\t25 STEP: ", test_mse_25)
f1, all_acc, acc_0, acc_1, acc_2, edges = eval_edges(
model, val_data, params)
print("Val Edge results:")
args.data_path,
'valid',
params,
num_in_path=False,
transpose_data=False,
max_len=40)
model = model_builder.build_model(params)
if args.mode == 'train':
with train_utils.build_writers(args.working_dir) as (train_writer,
val_writer):
train.train(model, train_data, val_data, params, train_writer,
val_writer)
elif args.mode == 'eval':
test_data = BasketballData(name,
args.data_path,
'test',
params,
num_in_path=False,
transpose_data=False)
test_cumulative_mse = evaluate.eval_forward_prediction(
model, test_data, 40, 9, params)
path = os.path.join(args.working_dir, args.error_out_name)
np.save(path, test_cumulative_mse.cpu().numpy())
test_mse_1 = test_cumulative_mse[0].item()
test_mse_5 = test_cumulative_mse[4].item()
test_mse_9 = test_cumulative_mse[8].item()
print("FORWARD PRED RESULTS:")
print("\t1 STEP: ", test_mse_1)
print("\t5 STEP: ", test_mse_5)
print("\t9 STEP: ", test_mse_9)