This paper introduces the "amplitude entropy". As for many of the other methods, this isn't any new entropy, but a new OutcomeSpace. Specifically, an input time series x is transformed as follows:
|z(𝑡)| = √︁(𝑥(𝑡)^2 + H{𝑥(𝑡)}^2)
where H{𝑥(𝑡)} is the Hilbert transform of the time series x(t).
They then bin the amplitude time series, normalise the histogram to form a a set of probabilities (i.e. use RelativeAmount), then compute Shannon entropy.
If possible, sketch out an implementation strategy
This should just be implemented as a new OutcomeSpace, and mention in the doctoring that it was originally used as part of the "amplitude entropy".
Describe the feature you'd like to have
This paper introduces the "amplitude entropy". As for many of the other methods, this isn't any new entropy, but a new
OutcomeSpace. Specifically, an input time seriesxis transformed as follows:|z(𝑡)| = √︁(𝑥(𝑡)^2 + H{𝑥(𝑡)}^2)
where H{𝑥(𝑡)} is the Hilbert transform of the time series x(t).
They then bin the amplitude time series, normalise the histogram to form a a set of probabilities (i.e. use
RelativeAmount), then compute Shannon entropy.If possible, sketch out an implementation strategy
This should just be implemented as a new
OutcomeSpace, and mention in the doctoring that it was originally used as part of the "amplitude entropy".