Influence-Maximization

Code and data for experiments with evolutionary influence maximization

Current development status

• Added extra argument to spread.MonteCarlo_simulation_max_hop, as it needed a reference to the pseudo-random number generator
• *args and **kargs adopted for passing fitness function as an argument
• BIG OOF, if we use multi-threaded stuff, we cannot guarantee replicability if the evaluator has pseudo-random number generation inside

Ideas for improvements

• Doina has C++ code that simulates spread much faster. Maybe it could be possible to create Python bindings for that code, passing it networkx graphs

Open questions tied to refactoring

evolutionaryalgorithm.py

• does it make sense to save the state of the population to different files?
• how should the function return the results?
• how should we structure the genotype? like, a set? should we keep the set coherent through all operators (operators always return correct sets), or should we instead just transform individuals to sets before evaluation (advantage: multi-threaded)?
• for some reason, preliminary experiments show no improvement when using several threads...to be investigated

Refactoring

• one file called "load" that contains everything related to loading functions
• one file called "spread" that contains all functions that evaluate influence spread over the network (simulation, lazy_fitness_evaluation, all approximations that do not really work)
• one file called "heuristics" that contains all heuristics that try to find influencial nodes in the network
• one file called "evolutionaryalgorithms" that contains all EAs (single and multi-objective)
• one file called "communities" that contains all methods related to community detection (we do not have stuff immediately available for networkx, but we might be able to find some library built on top of networkx)
• one file called "plots" to plot the graph in a nice way, highlighting the seed nodes (probably the way to go is to convert the graph from networkx to igraph and then use the nice plotting functions of igraph)
• one file called "threads" or "multithread" (check against existing modules), with the threadpool implementation by Alberto