Minimal example using DeepAREstimator

deepar_torch_minimal_example.py

from gluonts.dataset.repository.datasets import get_dataset
from gluonts.transform import ExpectedNumInstanceSampler
from gluonts.torch.model.deepar import DeepAREstimator
import torch

dataset = get_dataset("electricity")
context_length = 2 * 7 * 24
prediction_length = dataset.metadata.prediction_length

model = DeepAREstimator(
    prediction_length=prediction_length,
    context_length=context_length,
    freq='1H',
    hidden_size=100,
    train_sampler = ExpectedNumInstanceSampler(
        num_instances=1,
        min_future=prediction_length,
        min_past=context_length,
    ),
    batch_size=32,
    num_batches_per_epoch=50,
    trainer_kwargs={
        'max_epochs': 10, 
        'gpus': -1 if torch.cuda.is_available() else None,
    },
)

predictor = model.train(dataset.train, dataset.test, shuffle_buffer_length=10, num_workers=4)