parser.add_argument('--num-ref-points', type=int, default=128)
parser.add_argument('--classify-pertp', action='store_true')
args = parser.parse_args()
if __name__ == '__main__':
experiment_id = int(SystemRandom().random() * 100000)
print(args, experiment_id)
seed = args.seed
torch.manual_seed(seed)
np.random.seed(seed)
torch.cuda.manual_seed(seed)
random.seed(seed)
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
if args.dataset == 'physionet':
data_obj = utils.get_physionet_data(args, 'cpu', args.quantization)
elif args.dataset == 'mimiciii':
data_obj = utils.get_mimiciii_data(args)
train_loader = data_obj["train_dataloader"]
test_loader = data_obj["test_dataloader"]
val_loader = data_obj["val_dataloader"]
dim = data_obj["input_dim"]
if args.enc == 'enc_rnn3':
rec = models.enc_rnn3(dim,
torch.linspace(0, 1., 128),
args.latent_dim,
args.rec_hidden,
128,
learn_emb=args.learn_emb).to(device)
import numpy as np
from sklearn.model_selection import train_test_split
from utils import get_physionet_data
from models.ecg import ECGModel
# Load Physionet ECG data
_, data = get_physionet_data()
print('%i subjects loaded' % len(data))
# Train-validation split
ids = np.array(list(data.keys()))
train_ids, val_ids = train_test_split(ids, test_size=0.2, random_state=38)
train_data = [data[id] for id in train_ids]
val_data = [data[id] for id in val_ids]
# Train and save weights
trained_model_path = './trained_models/physionet'
model = ECGModel(latent_space_dim=8,
batch_size=64,
hidden_dim=128,
learning_rate=5e-4,
log_dir=trained_model_path)
model.fit(train_data, val_data)
model.save_weights(trained_model_path)