pyhctsa.operations.changepoint.stepdetect¶

pyhctsa.operations.changepoint.stepdetect(y, method='l1pwc', params=10)¶

Analysis of discrete steps in a time series.

Gives information about discrete steps in the signal, using the function l1pwc from Max A. Little’s step detection toolkit.

References

Parameters:¶

y : array-like¶

The input time series.

method : str¶

The step-detection method. Currently only l1pwc is supported.

params : float or str¶

The parameters for the given method used. For l1pwc this corresponds to lambda. Default is 10.

Returns:¶

Statistics on the output of the step-detection method, including the intervals between change points, the proportion of constant segments, the reduction in variance from removing the piece-wise constants, and stationarity in the occurrence of change points. The keys are E (objective per sample), s, lambdamax (scaled by 1/sqrt(N)), nsegments, rmsoff, rmsoffpstep, ratn12, diffn12 (the absolute difference between the numbers of change points in the two halves, as a proportion of the number of segments), pshort_3, meanstepintgt3 (in samples), cvstepint (std/mean of the step intervals) and medianstepint (in samples).

Return type:¶

dict