mojolearn.kpss_test#

mojolearn.kpss_test(y, d=0, D=0, s=0, pval_threshold=0.05, return_statistic=False)[source]#

KPSS stationarity test for a batch of series.

Differences each series (d simple and D seasonal differences) and runs the KPSS test, following cuml.tsa.stationarity.kpss_test. Series are given in columns, as in cuML. Computation is float32; float64 input is converted (and copied). Non-finite values raise an error naming the index, where cuML lets them reach the statistic. A constant series has statistic 0.0 and is judged stationary.

Parameters:

  • y (array-like of shape (n_obs, n_series) or (n_obs,)) – The series, one per column; 1-D is one series.

  • d (int, default 0) – Order of simple differencing.

  • D (int, default 0) – Order of seasonal differencing; D > 0 needs s >= 2. d + D is at most 2.

  • s (int, default 0) – Seasonal period.

  • pval_threshold (float, default 0.05) – Significance level.

  • return_statistic (bool, default False) – Also return the test statistic (cuML returns the flags only).

Returns:

  • stationary (Array of shape (n_series,), uint8) – 1 where the series is judged stationary after differencing, else 0.

  • statistic (Array of shape (n_series,), float32) – The KPSS statistic; returned only when return_statistic=True.

Examples

>>> import numpy as np, mojolearn
>>> y = np.random.default_rng(0).standard_normal((200, 4)).astype(np.float32)
>>> np.asarray(mojolearn.kpss_test(y)).shape
(4,)
On this page