quick start

Here is an example to train DLinear model in a long-term forecast settings, see ../examples/quickstart.py for more details

quick start

from torch_timeseries.dataset import ETTh1
from torch_timeseries.dataloader import StandardScaler, SlidingWindow, SlidingWindowTS
from torch_timeseries.model import DLinear
from torch.nn import MSELoss, L1Loss
from torch.optim import Adam
dataset = ETTh1('./data')
scaler = StandardScaler()
dataloader = SlidingWindowTS(dataset,
    window=96,
    horizon=1,
    steps=336,
    batch_size=32,
    train_ratio=0.7,
    val_ratio=0.2,
    scaler=scaler,
    )
model = DLinear(dataloader.window, dataloader.steps, dataset.num_features, individual= True)

optimizer = Adam(model.parameters())
loss_function = MSELoss()


# train
model.train()
for scaled_x, scaled_y, x, y, x_date_enc, y_date_enc in dataloader.train_loader:
    optimizer.zero_grad()

    scaled_x = scaled_x.float()
    scaled_y = scaled_y.float()
    scaled_pred_y = model(scaled_x)

    loss = loss_function(scaled_pred_y, scaled_y)
    loss.backward()
    optimizer.step()
    print(loss)