def __init__(self, H, dataset, transform=None):
super(CustomDataset, self).__init__('.', transform, None, None)
adj = H.to_sparse()
edge_index = adj._indices()
data_list = []
for i in range(0, len(dataset), 2):
data = Data(edge_index=edge_index)
data.x = dataset[i].t()
data.y = dataset[i + 1].t()
data_list.append(data)
self.data, self.slices = self.collate(data_list)
def __init__(self, H, dataset, x, fac_num, transform=None):
super(CustomDataset, self).__init__('.', transform, None, None)
adj = H.to_sparse()
edge_index = adj._indices()
data_list = []
for row in dataset:
data = Data(edge_index=edge_index)
row = torch.cat([row.unsqueeze(0).t(),
torch.zeros(fac_num, 1)],
dim=0)
data.x = row
data.y = x.t()
data_list.append(data)
self.data, self.slices = self.collate(data_list)
best_ckpt_path = os.path.join(ckpt_dir, 'best.ckpt')
last_ckpt_path = os.path.join(ckpt_dir, 'last.ckpt')
logger.setLevel(logging.INFO)
handler = logging.FileHandler(log_path, 'w')
handler.setLevel(logging.INFO)
formatter = logging.Formatter('%(asctime)s: %(message)s',
datefmt='%Y/%m/%d %H:%M:%S')
handler.setFormatter(formatter)
logger.addHandler(handler)
epochs = int(config.getfloat("hyperparameters", "max_epochs"))
learning_rate = config.getfloat("hyperparameters", "learning_rate")
weight_decay = config.getfloat("hyperparameters", "weight_decay")
dataset = Dataset(config)
model = init_model(config, device)
logger.info(model)
parameters = filter(lambda p: p.requires_grad, model.parameters())
#total_parameters = sum(p.numel() for p in parameters)
#logging.info("| there are totally {} trainable parameters".format(total_parameters))
for name, param in model.named_parameters():
print(name)
loss_func = nn.MSELoss()
optimizer = optim.Adam(parameters, lr=learning_rate)
least_loss = 9999999
log_path = os.path.join(ckpt_dir, 'train.log')
ckpt_path = os.path.join(ckpt_dir, 'best.ckpt')
logger.setLevel(logging.INFO)
handler = logging.FileHandler(log_path, 'w')
handler.setLevel(logging.INFO)
formatter = logging.Formatter('%(asctime)s: %(message)s',
datefmt='%Y/%m/%d %H:%M:%S')
handler.setFormatter(formatter)
logger.addHandler(handler)
epochs = int(config.getfloat("hyperparameters", "max_epochs"))
learning_rate = config.getfloat("hyperparameters", "learning_rate")
weight_decay = config.getfloat("hyperparameters", "weight_decay")
dataset = Dataset(config)
model = init_model(config, dataset.wordvec_weights, device)
dataset.wordvec_weights = None
logger.info(model)
parameters = filter(lambda p: p.requires_grad, model.parameters())
loss_func = nn.MSELoss()
optimizer = optim.Adam(parameters, lr=learning_rate)
least_loss = 9999999
model.to(device)
loss_func.to(device)
for epoch in range(epochs):
