time_windowed_features

time_windowed_features ¤

TimeWindowedFeatureTable ¤

TimeWindowedFeatureTable(
    dataframe: DataFrame, time_column: str = "systime"
)

Bases: Base

Build ML-ready feature tables from segmented timeseries data.

Takes the output of SegmentProcessor.apply_ranges and computes statistical metrics per UUID within fixed-size time windows (e.g. every 1 minute).

Methods: - compute_long: One row per (time_window, uuid, segment). Long format. - compute: One row per time_window with columns {uuid}__{metric}. Wide format.

compute_long `classmethod` ¤

compute_long(
    dataframe: DataFrame,
    window: str = "1min",
    time_column: str = "systime",
    uuid_column: str = "uuid",
    value_column: str = "value_double",
    segment_column: str | None = "segment_value",
    metrics: list[str] | None = None,
) -> pd.DataFrame

Compute statistical metrics per UUID per time window (long format).

Groups the input data into fixed-size time windows of window and computes numeric statistics for each (time_window, uuid, segment) group.

Parameters:

Name	Type	Description	Default
`dataframe`	`DataFrame`	Input DataFrame (typically output of SegmentProcessor.apply_ranges).	required
`window`	`str`	Pandas frequency string for the time window size (e.g. '1min', '30s', '5min'). `freq` is the deprecated name.	`'1min'`
`time_column`	`str`	Column containing timestamps.	`'systime'`
`uuid_column`	`str`	Column identifying each timeseries signal.	`'uuid'`
`value_column`	`str`	Column containing numeric values.	`'value_double'`
`segment_column`	`str \| None`	Column to sub-group by (e.g. 'segment_value'). Set to None to ignore segments entirely.	`'segment_value'`
`metrics`	`list[str] \| None`	Subset of metric names to compute. None uses all 19.	`None`

Windows are the bins of pd.Grouper(freq=window) (for sub-daily frequencies they start at midnight of the first day; on tz-aware data they are wall-clock windows in the data's zone). A window that spans a segment change gives one row per segment. Groups with fewer than 2 non-missing values are left out.

Returns:

Type	Description
`DataFrame`	DataFrame with columns [time_window, uuid, segment_value?,
`DataFrame`	sample_count, metric_1, ..., start, end, duration_seconds], one row
`DataFrame`	per (window, uuid, segment) sorted in that order. `start` is the
`DataFrame`	window start (= `time_window`), `end` the next window start.
`DataFrame`	Empty (with these columns) when no group has 2 values.

compute `classmethod` ¤

compute(
    dataframe: DataFrame,
    window: str = "1min",
    time_column: str = "systime",
    uuid_column: str = "uuid",
    value_column: str = "value_double",
    segment_column: str | None = "segment_value",
    metrics: list[str] | None = None,
    column_separator: str = "__",
) -> pd.DataFrame

Compute a wide-format feature table with one row per time window.

Each column is named {uuid}{separator}{metric} (e.g. temperature__mean). Windows where a UUID has insufficient data are filled with NaN.

Parameters:

Name	Type	Description	Default
`dataframe`	`DataFrame`	Input DataFrame (typically output of SegmentProcessor.apply_ranges).	required
`window`	`str`	Pandas frequency string for the time window size. `freq` is the deprecated name.	`'1min'`
`time_column`	`str`	Column containing timestamps.	`'systime'`
`uuid_column`	`str`	Column identifying each timeseries signal.	`'uuid'`
`value_column`	`str`	Column containing numeric values.	`'value_double'`
`segment_column`	`str \| None`	Column to sub-group by. Set to None to ignore segments.	`'segment_value'`
`metrics`	`list[str] \| None`	Subset of metric names to compute. None uses all 19.	`None`
`column_separator`	`str`	Separator between uuid and metric in wide column names.	`'__'`

Returns:

Type	Description
`DataFrame`	DataFrame with columns [time_window, segment_value?, {uuid}__{metric}, ...].

time_windowed_features