def test_distribution():
"""
Makes sure additional tensors can be accessed and have expected shapes
"""
prediction_length = ds_info.prediction_length
# todo adapt loader to anomaly detection use-case
batch_size = 2
num_samples = 3
estimator = DeepAREstimator(
freq=freq,
prediction_length=prediction_length,
trainer=Trainer(epochs=2, num_batches_per_epoch=1),
distr_output=StudentTOutput(),
batch_size=batch_size,
)
train_output = estimator.train_model(train_ds, test_ds)
training_data_loader = estimator.create_training_data_loader(
estimator.create_transformation().apply(train_ds))
seq_len = 2 * ds_info.prediction_length
for data_entry in islice(training_data_loader, 1):
input_names = get_hybrid_forward_input_names(
type(train_output.trained_net))
distr = train_output.trained_net.distribution(
*[data_entry[k] for k in input_names])
assert distr.sample(num_samples).shape == (
num_samples,
batch_size,
seq_len,
)
from gluonts.model.deepar import DeepAREstimator
from gluonts.trainer import Trainer
from gluonts.dataset.artificial import shuffle_testing_dataset
from gluonts.distribution.neg_binomial import NegativeBinomialOutput
BATCH_SIZE = 32
NUM_BATCH_PER_EPOCH = 50
n = 300
time_list = []
range_list = range(1, 30)
for c in range_list:
train_ds = shuffle_testing_dataset(n)
estimator = DeepAREstimator(
prediction_length=20,
freq="D",
)
transform = estimator.create_transformation()
loader = TrainDataLoader(
train_ds,
transform=transform,
batch_size=BATCH_SIZE,
ctx=mx.cpu(),
num_batches_per_epoch=NUM_BATCH_PER_EPOCH,
shuffle_buffer_length=c * BATCH_SIZE,
num_workers=2,
)
hit_list = [[0 for i in range(NUM_BATCH_PER_EPOCH)] for j in range(n)]
print(f"dataset size: {len(train_ds)}")
start_time = time.time()
count = 0
for batch in loader:
print(f"item id: {batch['item_id']}")
