inst/ctre-app/discussion.md

In strakaps/CTRE: Thresholding Bursty Time Series

The Mittag-Leffler distribution has a tail and a scale parameter, and setting tail=1 recovers the exponential distribution. Varying tail away from 1 (note that tail needs to be from the interval (0,1]) has two effects:

1. It models occasional very long quiet periods
2. It increases the frequency of events outside of quiet periods (makes the dynamics more "bursty").

The empirical datasets seem to prefer tail < 1 mostly because of the second effect. Repeated simulations show that the CTRM model predicts longer rests than are usually seen in the data. Hence the strength of the CTRM model lies in the modelling on short to medium scale event rather than long time scales.

strakaps/CTRE documentation built on Sept. 12, 2021, 6:12 p.m.