analyze(
y: ndarray,
x: ndarray | None,
blocks: Blocks,
washout: float,
design: Design,
) -> Analysis
The B - A coefficient of y on [1, z] or [1, z, X], its p-value and interval.
Refusals, in the order they are checked: those of make_frame,
then a block with no kept sample (empty_block), then an observed
assignment in the span of the covariates (collinear). The tie
tolerance is scaled by 1.4826 MAD of the kept target samples.
Examples:
Setting B adds 0.5 to a noisy target sampled once a minute:
>>> import numpy as np
>>> from tsdive.switchback import analyze, cut_blocks, make_design, setting
>>> design = make_design(24, seed=7)
>>> times = np.arange(24 * 60, dtype=float)
>>> blocks = cut_blocks(times, span=times.size, length=60.0)
>>> noise = np.random.default_rng(0).normal(0.0, 0.2, times.size)
>>> y = 0.5 * setting(blocks, design.observed) + noise
>>> result = analyze(y, None, blocks, washout=10.0, design=design)
>>> round(result.estimate, 2), round(result.lo, 2), round(result.hi, 2)
(0.49, 0.46, 0.52)
>>> round(result.p_value, 4), result.n_kept, result.reason
(0.001, 1200, '')