`Delay`#

Delay(duration) shifts every event's index by duration (in index units) and leaves its value unchanged: event (t, v) becomes (t + duration, v). It is the time-based counterpart of Lag, which shifts by a fixed number of events. On a regular grid the two coincide (duration = N * step equals Lag(N)); on an irregular feed Delay shifts by wall-time and Lag shifts by event count.

The shift is lossless, one output per input, order-preserving, and starts duration late (no NaN warmup). duration is numeric in index units (a millisecond index uses Delay(600_000) for a 10-minute delay); there is no calendar parsing.

Delay requires an explicit index (there is nothing to shift against without one); calling it on a bare array is a TypeError.

Parameters#

duration: the index offset to add to every event, numeric, in index units.

Limitations#

A Delay feeding a downstream merge (CombineLatest) inside a single live Pipeline emits the delayed event eagerly, so the merge would align it against a not-yet-advanced input. For an as-of alignment across a delay, apply Delay and CombineLatest as separate calls (batch CombineLatest sees all events), which is what forecast_pairs does. A fused-live-merge form needs a reorder buffer and is a planned follow-on.

Examples#

Delaying an irregular feed#

import numpy as np
import plotly.graph_objects as go
from screamer import Delay

idx = np.array([0, 7, 14, 21, 28], dtype=np.int64)
vals = np.array([1.0, 3.0, 2.0, 4.0, 3.5])
dv, di = Delay(5)(vals, idx)

fig = go.Figure()
fig.add_trace(go.Scatter(x=idx, y=vals, mode="lines+markers", name="input"))
fig.add_trace(go.Scatter(x=di, y=dv, mode="lines+markers", name="Delay(5)"))
fig.update_layout(title="Delay shifts the index by 5 units",
                  xaxis_title="index", yaxis_title="value",
                  margin=dict(l=20, r=20, t=50, b=20))
fig.show()

Reference#

A delay line from signal processing. Composes with CombineLatest to build an as-of-lagged join, and underpins forecast_pairs(duration=).

Delay#