Target Specification

PredictionWindowTargetSpec

from tempora.samplers import PredictionWindowTargetSpec

Targets that are generated according to a prediction or lookahead window.

PredictionWindowTargetSpec(
    window_len: int | float | dt.timedelta | np.timedelta64 | pd.Timedelta | Length,
    window_shift: int | float | dt.timedelta | np.timedelta64 | pd.Timedelta | Length | None = None,
    columns: list[str] | None = None,
    transform_spec: TransformSpec | LabelEncoder | EventEncoder | None = None
)

Parameters

Name	Description
window_len	Length of the prediction window.
window_shift	Optional shift between context window end and target window start.
columns	Optional target columns (defaults to all target-capable columns).
transform_spec	Optional TransformSpec, LabelEncoder, or EventEncoder for target-window transforms.

---

NextStepAheadTargetSpec

from tempora.samplers import NextStepAheadTargetSpec

Next-step ahead targets (requires the context window to be defined in rows).

NextStepAheadTargetSpec(
    columns: list[str] | None = None,
    transform_spec: TransformSpec | LabelEncoder | EventEncoder | None = None
)

Parameters

Name	Description
columns	Optional target columns.
transform_spec	Optional TransformSpec, LabelEncoder, or EventEncoder for target-window transforms.

---

ContextWindowTargetSpec

from tempora.samplers import ContextWindowTargetSpec

Targets that are generated according to the current context window.

ContextWindowTargetSpec(
    window_len: int | float | dt.timedelta | np.timedelta64 | pd.Timedelta | Length | None = None,
    window_shift: int | float | dt.timedelta | np.timedelta64 | pd.Timedelta | Length | None = None,
    columns: list[str] | None = None,
    transform_spec: TransformSpec | LabelEncoder | EventEncoder | None = None
)

Parameters

Name	Description
window_len	Optional target window length. Defaults to context length.
window_shift	Optional shift between context and target windows.
columns	Optional target columns.
transform_spec	Optional TransformSpec, LabelEncoder, or EventEncoder for target-window transforms.

---

SeriesTargetSpec

from tempora.samplers import SeriesTargetSpec

Target specification for full-series sampling.

SeriesTargetSpec(
    columns: list[str] | None = None,
    transform_spec: TransformSpec | LabelEncoder | EventEncoder | None = None
)

Parameters

Name	Description
columns	Optional subset of target columns. If all selected columns are non-temporal, the target for each sampled series is collapsed to a single row. If any selected column is temporal, the full series target window is returned.
transform_spec	Optional TransformSpec, LabelEncoder, or EventEncoder for target-window transforms. For non-temporal targets, the single-row collapse happens before the transform is applied.

Example: If each entity has a joined static class_label, then SeriesTargetSpec(columns=['class_label']) returns a scalar target for each sampled series.

---

TimeToEventTargetSpec

from tempora.samplers import TimeToEventTargetSpec

Time-to-event (TTE) targets.

TimeToEventTargetSpec(
    window_len: int | float | dt.timedelta | np.timedelta64 | pd.Timedelta | Length,
    window_shift: int | float | dt.timedelta | np.timedelta64 | pd.Timedelta | Length | None = None,
    columns: list[str] | None = None,
    censored: bool = False
)

Parameters

Name	Description
window_len	Window length for event detection.
window_shift	Optional shift between context and target windows.
columns	Optional target columns.
censored	Whether to treat missing events as censored.