def test_ExpectedNumInstanceSampler():
N = 6
train_length = 2
pred_length = 1
ds = make_dataset(N, train_length)
t = transform.Chain(trans=[
transform.InstanceSplitter(
target_field=FieldName.TARGET,
is_pad_field=FieldName.IS_PAD,
start_field=FieldName.START,
forecast_start_field=FieldName.FORECAST_START,
instance_sampler=transform.ExpectedNumInstanceSampler(
num_instances=4, min_future=pred_length),
past_length=train_length,
future_length=pred_length,
)
])
assert_serializable(t)
scale_hist = ScaleHistogram()
repetition = 2
for i in range(repetition):
for data in t(iter(ds), is_train=True):
target_values = data["past_target"]
# for simplicity, discard values that are zeros to avoid confusion with padding
target_values = target_values[target_values > 0]
scale_hist.add(target_values)
expected_values = {i: 2**i * repetition for i in range(1, N)}
assert expected_values == scale_hist.bin_counts
def test_dataset_statistics(self) -> None:
n = 2
T = 10
# use integers to avoid float conversion that can fail comparison
np.random.seed(0)
targets = np.random.randint(0, 10, (n, T))
scale_histogram = ScaleHistogram()
for i in range(n):
scale_histogram.add(targets[i, :])
scale_histogram.add([])
expected = DatasetStatistics(
integer_dataset=True,
num_time_series=n + 1,
num_time_observations=targets.size,
mean_target_length=T * 2 / 3,
min_target=targets.min(),
mean_target=targets.mean(),
mean_abs_target=targets.mean(),
max_target=targets.max(),
cats=[{0}, {1, 2}],
num_dynamic_feat=2,
num_missing_values=0,
scale_histogram=scale_histogram,
)
# FIXME: the cast below is a hack to make mypy happy
timeseries = cast(
Dataset,
[
make_time_series(
target=targets[0, :], cat=[0, 1], num_dynamic_feat=2
),
make_time_series(
target=targets[1, :], cat=[0, 2], num_dynamic_feat=2
),
make_time_series(
target=np.array([]), cat=[0, 2], num_dynamic_feat=2
),
],
)
found = calculate_dataset_statistics(timeseries)
assert expected == found
def test_BucketInstanceSampler():
N = 6
train_length = 2
pred_length = 1
ds = make_dataset(N, train_length)
dataset_stats = calculate_dataset_statistics(ds)
t = transform.Chain(
trans=[
transform.InstanceSplitter(
target_field=transform.FieldName.TARGET,
is_pad_field=transform.FieldName.IS_PAD,
start_field=transform.FieldName.START,
forecast_start_field=transform.FieldName.FORECAST_START,
train_sampler=transform.BucketInstanceSampler(
dataset_stats.scale_histogram
),
past_length=train_length,
future_length=pred_length,
pick_incomplete=True,
)
]
)
assert_serializable(t)
scale_hist = ScaleHistogram()
repetition = 200
for i in range(repetition):
for data in t(iter(ds), is_train=True):
target_values = data["past_target"]
# for simplicity, discard values that are zeros to avoid confusion with padding
target_values = target_values[target_values > 0]
scale_hist.add(target_values)
expected_values = {i: repetition for i in range(1, N)}
found_values = scale_hist.bin_counts
for i in range(1, N):
assert abs(
expected_values[i] - found_values[i] < expected_values[i] * 0.3
)
def test_dataset_statistics(self) -> None:
num_time_series = 3
num_time_observations = 10
num_feat_dynamic_real = 2
num_past_feat_dynamic_real = 3
num_feat_dynamic_cat = 2
num_missing_values = 0
# use integers to avoid float conversion that can fail comparison
np.random.seed(0)
targets = np.random.randint(
0, 10, (num_time_series - 1, num_time_observations)
)
scale_histogram = ScaleHistogram()
for i in range(num_time_series - 1):
scale_histogram.add(targets[i, :])
scale_histogram.add([])
expected = DatasetStatistics(
integer_dataset=True,
num_time_series=num_time_series, # includes empty array
num_time_observations=targets.size,
mean_target_length=num_time_observations
* (num_time_series - 1)
/ num_time_series,
max_target_length=num_time_observations,
min_target=targets.min(),
mean_target=targets.mean(),
mean_abs_target=targets.mean(),
max_target=targets.max(),
feat_static_real=[{0.1}, {0.2, 0.3}],
feat_static_cat=[{1}, {2, 3}],
num_feat_dynamic_real=num_feat_dynamic_real,
num_past_feat_dynamic_real=num_past_feat_dynamic_real,
num_feat_dynamic_cat=num_feat_dynamic_cat,
num_missing_values=num_missing_values,
scale_histogram=scale_histogram,
)
# FIXME: the cast below is a hack to make mypy happy
timeseries = cast(
Dataset,
[
make_time_series(
target=targets[0, :],
feat_static_cat=[1, 2],
feat_static_real=[0.1, 0.2],
num_feat_dynamic_cat=num_feat_dynamic_cat,
num_feat_dynamic_real=num_feat_dynamic_real,
num_past_feat_dynamic_real=num_past_feat_dynamic_real,
),
make_time_series(
target=targets[1, :],
feat_static_cat=[1, 3],
feat_static_real=[0.1, 0.3],
num_feat_dynamic_cat=num_feat_dynamic_cat,
num_feat_dynamic_real=num_feat_dynamic_real,
num_past_feat_dynamic_real=num_past_feat_dynamic_real,
),
make_time_series(
target=np.array([]),
feat_static_cat=[1, 3],
feat_static_real=[0.1, 0.3],
num_feat_dynamic_cat=num_feat_dynamic_cat,
num_feat_dynamic_real=num_feat_dynamic_real,
num_past_feat_dynamic_real=num_past_feat_dynamic_real,
),
],
)
found = calculate_dataset_statistics(timeseries)
assert expected == found