Higher Order Probabilistic Analysis of Network Trajectories of Intelligent Agents in Thespian.

Intelligent agents (IAs) are autonomous pieces of software designed to be deployed to and operate on infrastructure, whether physical or digital, and perform various tasks on them ranging from integrity check and structure discovery to functionality monitoring and information collection. Their task performing capability has been considerably increased with the recent advent of machine learning. An important parameter of IAs operating on networks such as the Web is their trajectory, which in many engineering scenarios depends heavily on random outcomes taking place at each vertex visited by the IA. In order to study the probabilistic properties of the trajectory length, said outcomes instead of being modeled or simulated are computed as the result as a game taking place between the vertex and the IA, developed in the Thespian framework for Python, and the vertex. The latter selects a random but fixed strategy, whereas the IA can adapt to learn this strategy either by observing the entropy of the choices of its opponent. If IA loses, then it backtracks, otherwise it chooses its next destination with a preferential attachment scheme. The mean, variance, skewness, and kurtosis of the trajectories of IAs operating on three scale-free graphs generated by NetworkX. Emphasis was placed on proper Pythonic code as many major Python modules such as threads, for generating game instances, Counter instances from collections to keep track of player choices, and functools for map/reduce functionality. Results indicate that IAs learning the opponent strategy have longer and richer network trajectories in terms of vertices, indicating the importance of learning.