Ctrl+K

detect_cusum#

pyCGM2.Signal.detector.detect_cusum(x: ndarray, threshold: int = 1, drift: int = 0, ending: bool = False, show: bool = True, ax: Axes | None = None)#

Cumulative sum algorithm (CUSUM) to detect abrupt changes in data.

Parameters:

  • x (np.ndarray) – Data in which to detect changes.

  • threshold (int, optional) – Amplitude threshold for the change in the data. Default is 1.

  • drift (int, optional) – Drift term that prevents any change in the absence of change. Default is 0.

  • ending (bool, optional) – If True, estimates when the change ends. If False, does not. Default is False.

  • show (bool, optional) – If True, plots data in matplotlib figure. If False, does not. Default is True.

  • ax (plt.Axes, optional) – A matplotlib Axes instance for the plot. Default is None.

Returns:

tuple – A tuple containing: - ta (np.ndarray): Alarm time (index of when the change was detected). - tai (np.ndarray): Index of when the change started. - taf (np.ndarray): Index of when the change ended (if ending is True). - amp (np.ndarray): Amplitude of changes (if ending is True).

Notes

  • Tuning of the CUSUM algorithm according to Gustafsson (2000): Start with a very large threshold. Choose drift to half the expected change, then adjust so that g = 0 more than 50% of the time. Set the threshold for the desired false alarm rate or detection delay. Decrease drift for faster detection, increase for fewer false alarms.

  • By default, repeated sequential changes are not deleted. Set ending to True to delete them.

  • See the referenced IPython Notebook for more information.

References

  • Gustafsson (2000) Adaptive Filtering and Change Detection.

  • [CUSUM Notebook](demotu/detecta)

Examples

>>> x = np.random.randn(300)/5
>>> x[100:200] += np.arange(0, 4, 4/100)
>>> ta, tai, taf, amp = detect_cusum(x, 2, .02, True, True)
>>> x = np.random.randn(300)
>>> x[100:200] += 6
>>> detect_cusum(x, 4, 1.5, True, True)
>>> x = 2*np.sin(2*np.pi*np.arange(0, 3, .01))
>>> ta, tai, taf, amp = detect_cusum(x, 1, .05, True, True)
Version history:
On this page
Show Source