Influence-Maximization

Code and data for experiments with evolutionary influence maximization

This repository contains source code and some experimental scripts for experimental setups. The inputs, outputs, plots and other experiment-specific files are contained in another subrepository: experiments.

Influence-Maximization repository is organized as follows:

• src: source code containing code of the implemented features
• src_OLD: old source code in case it comes at hand
• tests: some tests needed to check whether some implemented functions behave properly
• experiments: subrepository containing inputs and outputs of each experimental setup
• datasets: contains script that downloads datasets

The src module contains the following submodules:

• ea: code related to each of the components of the evolutionary algorithm
• spread: implementations of fitness functions approximations
• spread_pyx: cython code and compilation script
• nodes_filtering: implementation of different nodes filtering strategies
• additional_experimental_scripts: experimental setup scripts of all old experiments

and files:

• evolutionary_algorithm_exec: script containing the execution pipeline: graph loading, nodes filtering, smart initialization, execution of the ea algorithm with appropriate parameters, output logging..
• heuristics: implementation of other heuristics we compare with
• multi_armed_bandit: implementation of the UCB1 algorithm for multi-armed-bandit problem (used for dynamic mutation selection)
• smart_initialization: implementation of different smart initialization techniques
• utils: mix of helpful functions
• experiments: script used for execution of multiple experiments
• load.py: graph loading code
• mo_evolutionaryalgorithm: multi-objective old evolutionary algorithm code

To run the whole pipeline:

cd Influence-Maximization; python -m src.evolutionary_algorithm_exec --config_file src/example_experimental_setup.json

You need to specify all the hyperparameters of the algorithm in a .json config file and pass it as a parameter of the script. The results are stored in a directory specified as a script argument (--out_dir).

Limitations

Pay attention: when you change the mutation_operator parameter, currently you need to add a decorator @inspyred.ec.variators.mutator to the selected mutation function in src/ea/mutators.py.

Python and libraries versions currently used:

• python 3.6.8
• inspyred 1.0.1
• networkx 2.3
• numpy 1.16.3
• cython 0.29.13
• node2vec 0.3.1
• gensim 3.8.1
• pandas 0.24.2