In modelling a sequence of adaptive choices by an intelligent agent (e.g. places visited, web sites browsed), memory-less random walks are unsuitable, because of the formation of agent habits and preferences.
Often these choices are only partially observed, and report times are sporadic and bursty, in contrast to regular or exponentially spaced times in classical models.
The FRACTALRABBIT stochastic mobility simulator creates realistic synthetic sporadic waypoint data sets. It consist of three tiers, each based on new stochastic models:
(1) An Agoraphobic Point Process generates a set V of space points, whose limit is a random fractal, representing sites that could be visited.
(2) A Retro-preferential Process generates a trajectory X through V , with strategic homing and self-reinforcing site ﬁdelity as observed in human/animal behavior.
(3) A Sporadic Reporting Process models time points T at which the trajectory X is observed, with bursts of reports and heavy tailed inter-event times.
FRACTALRABBIT can be used to test algorithms applicable to sporadic waypoint data, such as (1) co-travel mining, (2) anomaly detection, and (3) extraction of maximal self-consistent subsets of corrupted data.
Reference: R. W. R. Darling, "Retro-preferential Stochastic Mobility Models on Random Fractals Under Sporadic Observations", DOI: 10.13140/RG.2.2.15267.40489, 2018
- Bugs and feature requests
- Copyright and license
Java version runs from the command line:
java -jar fractalrabbit.jar parameters.csv
An example of the parameters.csv file is provided in the resources folder. Change it to suit your modelling needs. It permits multiple travellers to follow the same trajectory asynchronously.
- Have a bug or a feature request? Contact Github user bbux-atg
- See Wiki.
- New implementations of the three underlying models described in the technical report are welcome.
R. W. R. Darling http://probabilist.us Github: probabilist-us
Apache License 2.0