Appl. of the Minimal Dominating Set for Influence Max. in Multilayer Networks

A repository with a source code for the paper: https://arxiv.org/abs/2502.15236

• Authors: Michał Czuba(¶†), Mingshan Jia(†), Piotr Bródka(¶†), Katarzyna Musial(†)
• Affiliation:
  (¶) WUST, Wrocław, Lower Silesia, Poland
  (†) UTS, Sydney, NSW, Australia

Configuration of the runtime

First, initialise the enviornment:

conda env create -f env/conda.yaml
conda activate infmax-mds-ltm-mln
python -m ipykernel install --user --name=infmax-mds-ltm-mln

To use scripts which produce analysis, install the source code:

pip install -e .

Data

Dataset is stored with DVC. Thus, to obtain it you have to access a Google Drive. Please send a request via e-mail (michal.czuba@pwr.edu.pl) to have it granted. Then, execute in shell: dvc pull. zip files are large and we recommend to pull them only if necessary.

Structure of the repository

.
├── README.md
├── data
│   ├── brute_ds            -> a brute force DS finder
│   ├── networks            -> networks used in exmeriments
│   ├── processed_results
│   ├── raw_results
│   └── test                -> examplary results of the simulator used in the E2E test
├── env                     -> a definition of the runtime environment
├── scripts
│   ├── analysis
│   └── configs             -> exemplary configuration files
├── src                     -> scripts to execute experiments and process the results
├── run_experiments.py      -> an entrypoint to trigger the pipeline to evaluate MDS in InfMax
├── test_reproducibility.py -> E2E test to prove that results can be repeated

Series of the results

• batch_1 real-world, g-mds, AND
• batch_2 real-world, g-mds, OR
• batch_3 artificial, g-mds, AND
• batch_4 artificial, g-mds, OR
• batch_5 timik1q2009, g-mds, AND
• batch_6 timik1q2009, g-mds, OR
• batch_7 real-world, li-mds, AND
• batch_8 real-world, li-mds, OR
• batch_9 artificial, li-mds, AND
• batch_10 artificial, li-mds, OR
• batch_11 timik1q2009, li-mds, AND
• batch_12 timik1q2009, li-mds, OR

Running the pipeline

To run experiments execute: python run_experiments.py <config file>. See example_config.yaml for inspirations. As a result, for each repetition of the cartesian product computed for the provided parameters, a csv file will be obtained with following columns:

{
    seed_ids: str           # IDs of actors that were seeds aggr. into string (sep. by ;)
    gain: float             # gain* obtained using this seed set
    simulation_length: int  # nb. of simulation steps
    seed_nb: int            # nb. of actors that were seeds
    exposed_nb: int         # nb. of active actors at the end of the simulation
    unexposed_nb: int       # nb. of actors that remained inactive
    expositons_rec: str     # record of new activations in each epoch aggr. into string (sep. by ;)
    network: str            # network's name
    protocol: str           # protocols's name
    seed_budget: float      # value of the maximal seed budget
    mi_value: float         # value of the threshold
    ss_method: str          # seed selection method's name
}

* Gain is the percentage of the non-initially seeded population that became exposed during the simulation: (exposed_nb - seed_nb) / (total_actor_nb - seed_nb) * 100%

The simulator will also save provided configuraiton file, rankings of actors used in computations, and detailed logs of evaluated cases whose index divided modulo by full_output_frequency equals 0.

Results reproducibility

Results are supposed to be fully reproducable. There is a test for that: test_reproducibility.py.

Obtaining analysis of results

To process raw results please execute scripts in scripts/analysis directory in the order as depicted in a following tree. Please note, that names of scripts reflect names of genreated files under data/processed_results:

.
├── distr_expos.ipynb
├── quantitative_comparison.py
│   ├── effectiveness_heatmaps.py
│   └── profile_reports.py
├── metrics.py
├── similarities_mds.py
├── similarities_seeds.py
├── visualisations_mds.py
└── mds_algos_comparison.py