def train(datase, args):
if args.dataset == 'AE':
args.data_path = '../dataset/energydata_complete.csv'
args.inputs_index = ['Windspeed']
if 'PM' in args.dataset:
args.data_path = '../dataset/{}20100101_20151231.csv'.format(
args.dataset)
args.inputs_index = ['Iws']
args.num_inputs = len(args.inputs_index)
pred_length_list = [1] ## ranging from ten minutes to 60 minutes
dropout_list = [0, 0.1]
for dropout, pred_length in itertools.product(dropout_list,
pred_length_list):
args.dropout = dropout
args.pred_length = pred_length
args.train_length = args.timesteps + args.pred_length
model = DMDNN_model(args, device)
load_path = os.path.join(model.args.dataset, model.model_dir,
model.args.save_model_path)
if not os.path.exists(load_path):
print('load path {} does not exist, training from scratch'.format(
load_path))
train_losses = model.train()
model.save_model()
else:
print('load path exists, loading...')
model.load_model()
#### validation on the validation set##########
real_y, final_generation = model.CRPS_generation(
args.num_sampling, args.num_experiments, mode='validation'
) ## shape of num_experiments, num_sampling, N, dim
CRPS_score = CRPS(real_y, final_generation, average=args.average)
model.save_score(CRPS_score, mode='validation', metric='CRPS')
real_y, final_generation = model.CRPS_generation(
args.num_sampling, args.num_experiments, mode='validation'
) ## shape of num_experiments, num_sampling, N, dim
RMSE_score = RMSE(real_y, final_generation, average=args.average)
model.save_score(RMSE_score, mode='validation', metric='RMSE')
# #test on the test dataset
real_y, final_generation = model.CRPS_generation(
args.num_sampling, args.num_experiments,
mode='test') ## shape of num_experiments, num_sampling, N, dim
CRPS_score = CRPS(real_y, final_generation, average=args.average)
model.save_score(CRPS_score, mode='test', metric='CRPS')
real_y, final_generation = model.CRPS_generation(
args.num_sampling, args.num_experiments,
mode='test') ## shape of num_experiments, num_sampling, N, dim
RMSE_score = RMSE(real_y, final_generation, average=args.average)
model.save_score(RMSE_score, mode='test', metric='RMSE')
def train(dataset, args):
if args.dataset == 'AE':
args.data_path = '../dataset/energydata_complete.csv'
args.inputs_index = ['Windspeed']
if 'PM' in args.dataset:
args.data_path = '../dataset/{}20100101_20151231.csv'.format(
args.dataset)
args.inputs_index = ['Iws']
pred_length_list = [1] ## ranging from ten minutes to 60 minutes
scores = []
for pred_length in pred_length_list:
args.pred_length = pred_length
args.train_length = args.timesteps + args.pred_length
model = MPersistent(args, device)
### test stage###
real_y, final_generation = model.pred_eval('test')
# ## if not denormalized, do this part
# real_y=real_y*model.dataset.std+model.dataset.mean
# final_generation=final_generation*model.dataset.std+model.dataset.mean
# path=model.args.test_path
# save_path=os.path.join(model.args.dataset, model.model_dir, path)
# model.save_generation(real_y, final_generation, save_path)
MAE_score = MAE(real_y, final_generation, average=args.average)
scores.append(MAE_score)
model.save_score(MAE_score, mode='test', metric='MAE')
RMSE_score = RMSE(real_y, final_generation, average=args.average)
scores.append(RMSE_score)
model.save_score(RMSE_score, mode='test', metric='RMSE')
print(scores)
def train(dataset, args):
if args.dataset == 'AE':
args.data_path = '../dataset/energydata_complete.csv'
args.inputs_index = ['Windspeed']
if 'PM' in args.dataset:
args.data_path = '../dataset/{}20100101_20151231.csv'.format(
args.dataset)
args.inputs_index = ['Iws']
pred_length_list = [1] ## ranging from ten minutes to 60 minutes
for pred_length in pred_length_list:
args.pred_length = pred_length
args.train_length = args.timesteps + args.pred_length
model = GP(args, device)
### test stage###
real_y, final_generation = model.pred_eval(args.num_sampling,
args.num_experiments,
'test')
MAE_score = CRPS(real_y, final_generation, average=args.average)
model.save_score(MAE_score, mode='test', metric='CRPS')
print(MAE_score)
RMSE_score = RMSE(real_y, final_generation, average=args.average)
model.save_score(RMSE_score, mode='test', metric='RMSE')
print(RMSE_score)
print('load path {} does not exist, training from scratch'.format(
load_path))
train_losses = model.train()
model.save_model()
else:
print('load path exists, loading...')
model.load_model()
#### validation on the validation set##########
real_y, final_prediction = model.CRPS_prediction(
args.num_sampling, args.num_experiments, mode='validation'
) ## shape of num_experiments, num_sampling, N, dim
CRPS_score = CRPS(real_y, final_prediction, average=args.average)
model.save_score(CRPS_score, mode='validation', metric='CRPS')
real_y, final_prediction = model.CRPS_prediction(
args.num_sampling, args.num_experiments, mode='validation'
) ## shape of num_experiments, num_sampling, N, dim
RMSE_score = RMSE(real_y, final_prediction, average=args.average)
model.save_score(RMSE_score, mode='validation', metric='RMSE')
# #test on the test dataset
real_y, final_prediction = model.CRPS_prediction(
args.num_sampling, args.num_experiments,
mode='test') ## shape of num_experiments, num_sampling, N, dim
CRPS_score = CRPS(real_y, final_prediction, average=args.average)
model.save_score(CRPS_score, mode='test', metric='CRPS')
real_y, final_prediction = model.CRPS_prediction(
args.num_sampling, args.num_experiments,
mode='test') ## shape of num_experiments, num_sampling, N, dim
RMSE_score = RMSE(real_y, final_prediction, average=args.average)
model.save_score(RMSE_score, mode='test', metric='RMSE')
def train(dataset, args):
if args.dataset == 'AE':
args.data_path = '../dataset/energydata_complete.csv'
args.inputs_index = ['Windspeed']
if 'PM' in args.dataset:
args.data_path = '../dataset/{}20100101_20151231.csv'.format(
args.dataset)
args.inputs_index = ['Iws']
args.num_inputs = len(args.inputs_index)
pred_length_list = [1] ## ranging from ten minutes to 60 minutes
dropout_list = [0, 0.1]
for dropout, pred_length in itertools.product(dropout_list,
pred_length_list):
args.dropout = dropout
args.pred_length = pred_length
args.train_length = args.timesteps + args.pred_length
###training stage####
model = PANN(args, device)
load_path = os.path.join(model.args.dataset, model.model_dir,
model.args.save_model_path)
if not os.path.exists(load_path):
print('load path does not exist, training from scratch')
train_losses = model.train()
model.save_model()
else:
print('load path exists, loading...')
model.load_model()
### validation stage###
real_y, final_generation = model.pred_eval('validation')
# ## if not denormalized, do this part
# real_y=real_y*model.dataset.std+model.dataset.mean
# final_generation=final_generation*model.dataset.std+model.dataset.mean
# path=model.args.validation_path
# save_path=os.path.join(model.args.dataset, model.model_dir, path)
# model.save_generation(real_y, final_generation, save_path)
MAE_score = MAE(real_y, final_generation, average=args.average)
model.save_score(MAE_score, mode='validation', metric='MAE')
RMSE_score = RMSE(real_y, final_generation, average=args.average)
model.save_score(RMSE_score, mode='validation', metric='RMSE')
### test stage###
real_y, final_generation = model.pred_eval('test')
# ## if not denormalized, do this part
# real_y=real_y*model.dataset.std+model.dataset.mean
# final_generation=final_generation*model.dataset.std+model.dataset.mean
# path=model.args.test_path
# save_path=os.path.join(model.args.dataset, model.model_dir, path)
# model.save_generation(real_y, final_generation, save_path)
MAE_score = MAE(real_y, final_generation, average=args.average)
model.save_score(MAE_score, mode='test', metric='MAE')
RMSE_score = RMSE(real_y, final_generation, average=args.average)
model.save_score(RMSE_score, mode='test', metric='RMSE')