pyhctsa.operations.stationarity.local_global

pyhctsa.operations.stationarity.local_global(y, subset_how='l', n=None, random_seed=None)

Compare local statistics to global statistics of a time series.

Parameters:
y : array-like

The time series to analyse.

subset_how : str, optional

The method to select the local subset of time series:

  • ’l’: the first n points in a time series

  • ’p’: an initial proportion of the full time series (round(N*n) points, halves rounded up)

  • ’unicg’: n evenly-spaced points throughout the time series

  • ’randcg’: n randomly-chosen points from the time series (chosen with replacement)

Default is 'l'.

n : int or float, optional

The parameter for the method specified by subset_how.

Default None is 100 samples or 0.1 (10% of time series length) if proportion.

random_seed : int, str or None, optional

Seed for the ‘randcg’ option: an integer, None or 'default' for seed 0, or 'none' for a seed drawn from NumPy’s global random state. The points come from the portable generator pyhctsa.robust.bf_random() (uniform on the indices, with repeats), so they are hctsa’s for the same seed. Default None is seed 0.

Returns:

A dictionary containing various statistical measures comparing the subset to the full time series: absmean, std, median, iqr, skewness, kurtosis, ac1 (the last four are the absolute relative deviation of the subset statistic from the full-series one, NaN when the full-series statistic is zero).

Return type:

dict