Result Types

OfflineChangePointResult

Returned by all offline detectors. Contains detected breakpoints, optional scores and segment statistics, and rich diagnostics.

Properties

Property	Type	Description
breakpoints	list[int]	Breakpoint indices (strictly increasing, final element = n)
change_points	list[int]	Change point indices (breakpoints excluding the terminal n)
scores	list[float] | None	Per-change-point scores (length = len(change_points), if available)
segments	list[SegmentStats] | None	Per-segment statistics (length = len(breakpoints))
diagnostics	Diagnostics	Rich diagnostics metadata

Methods

to_json() -> str

Serialize the result to a JSON string following the versioned contract.

• Current writer version: schema marker 1

• Unknown fields are preserved during round-trip

• Optional build provenance is surfaced under diagnostics.build when available

from_json(payload: str) -> OfflineChangePointResult (static)

Deserialize a result from a JSON string.

• Accepts schema markers in the supported window (currently 1..=2)

• Raises: ValueError on unsupported schema version or structural validation failure

plot(values=None, *, ax=None, title=None, breakpoint_color="crimson", breakpoint_style="--", line_width=1.5, show_legend=True) -> Any

Plot the signal with detected breakpoints overlaid.

Parameter	Type	Default	Description
values	array | None	None	Signal values. Required if segments is None
ax	matplotlib.Axes | None	None	Axes to plot on. Creates new figure if None
title	str | None	None	Plot title
breakpoint_color	str	"crimson"	Color for breakpoint lines
breakpoint_style	str	"--"	Line style for breakpoints
line_width	float	1.5	Line width
show_legend	bool	True	Whether to show legend

• Returns: matplotlib Figure

• Raises: ImportError if matplotlib is not installed

Note

plot(ax=...) is supported only for univariate data (diagnostics.d == 1).

Example

result = cpd.Pelt(model="l2").fit(x).predict(n_bkps=2)

# Inspect results
print(result.breakpoints)    # [50, 100, 150]
print(result.change_points)  # [50, 100]

# Serialize round-trip
payload = result.to_json()
restored = cpd.OfflineChangePointResult.from_json(payload)
assert restored.breakpoints == result.breakpoints

# Plot
fig = result.plot(x, title="Detected breakpoints")

---

OnlineStepResult

Emitted by online detectors for each processed observation.

Properties

Property	Type	Description
t	int	Time index (0-based observation count)
p_change	float	Posterior probability of a change at this step
alert	bool	Whether the alert policy fired
alert_reason	str | None	Reason the alert fired (e.g. "threshold")
run_length_mode	int	Mode of the run-length posterior
run_length_mean	float	Mean of the run-length posterior
processing_latency_us	int | None	Processing latency in microseconds (if measured)

Example

bocpd = cpd.Bocpd(model="gaussian_nig", hazard=1/200)
step = bocpd.update(3.5)

print(step.t)          # 0
print(step.p_change)   # 0.005 (low for first observation)
print(step.alert)      # False