datasets

datasets

Synthetic timeseries generators for trying out ts-shape.

These helpers produce DataFrames in the standard ts-shape schema (systime, uuid, value_*, is_delta) so any detector, transform, or statistic can be exercised without real data or loaders::

import ts_shape
df = ts_shape.make_timeseries(["sensor:temp"], n_outliers=4)
ts_shape.OutlierDetectionEvents(df, value_column="value_double") \
    .detect_outliers_zscore()

make_timeseries

make_timeseries(
    uuids: Sequence[str] = ("sensor:signal",),
    *,
    n_points: int = 1000,
    freq: str = "30s",
    start: str = "2025-01-01 00:00:00",
    baseline: float = 100.0,
    noise: float = 1.0,
    drift: float = 0.0,
    n_outliers: int = 0,
    value_column: str = "value_double",
    seed: int | None = 42
) -> pd.DataFrame

Generate a standard-schema synthetic timeseries DataFrame.

Parameters:

Name	Type	Description	Default
uuids	Sequence[str]	Signal identifiers; one block of n_points rows per uuid.	('sensor:signal',)
n_points	int	Number of samples generated per uuid.	1000
freq	str	Pandas offset alias for the sampling interval (e.g. "30s").	'30s'
start	str	Timestamp of the first sample.	'2025-01-01 00:00:00'
baseline	float	Mean level of the generated signal.	100.0
noise	float	Standard deviation of the Gaussian noise added to the signal.	1.0
drift	float	Total linear drift applied across the series (tool-wear style).	0.0
n_outliers	int	Number of large spike outliers injected per uuid.	0
value_column	str	Which value column to populate -- one of value_double, value_integer or value_bool.	'value_double'
seed	int | None	Seed for reproducibility; pass None for fresh randomness.	42

Returns:

Type	Description
DataFrame	DataFrame with columns systime, uuid, value_bool,
DataFrame	value_integer, value_double, value_string and is_delta.

Raises:

Type	Description
ValueError	If value_column is not a supported value column, or n_points is not positive.

make_id_signal

make_id_signal(
    uuid: str = "object:id",
    values: Sequence[str] = ("A", "B", "C"),
    *,
    hold: int = 10,
    freq: str = "30s",
    start: str = "2025-01-01 00:00:00",
    value_column: str = "value_string",
    source_uuid: str | None = None
) -> pd.DataFrame

Generate a categorical identifier signal in the standard schema.

Each value in values is held for hold consecutive samples, so the signal looks like a real batch / serial / coil / recipe id stream that changes over time. Feed it to :mod:ts_shape.eventlog.objects to extract object instances.

Parameters:

Name	Type	Description	Default
uuid	str	Signal identifier.	'object:id'
values	Sequence[str]	Ordered id values; each held for hold samples.	('A', 'B', 'C')
hold	int	Samples each value persists before the next one starts.	10
freq	str	Pandas offset alias for the sampling interval.	'30s'
start	str	Timestamp of the first sample.	'2025-01-01 00:00:00'
value_column	str	Which column carries the id (value_string or value_integer).	'value_string'
source_uuid	str | None	Optional source_uuid to stamp on every row.	None

Returns:

Type	Description
DataFrame	DataFrame with the standard ts-shape columns, value_column filled.