Author: Daniele Zambon,
Affiliation: Università della Svizzera italiana
eMail: daniele.zambon@usi.ch
Last Update: 31/01/2020

Change-Point Methods on a Sequence of Graphs

@article{zambon2019change,
  title={Change-Point Methods on a Sequence of Graphs},
  author={Daniele Zambon, Cesare Alippi and Lorenzo Livi},
  year={2019},
  doi={10.1109/TSP.2019.2953596},
  journal={IEEE Transactions on Signal Processing},
}

Description

• run_simulation.py is the main script that runs a repeated experiement (with the same settings). It adopts the framework cdg.simulations in cdg.
• settings.py contains info related to the datasets and declares the predefined experimental settings.
• config.py define parameters like: seed, folder of the results and significance level of the tests.
• red_button.py is a wrapper that allows to run multiple experiments.

Run experiments

Get the data

Delaunay graphs can be directly generated by cdg. The procedure is the following. Check in the file settings.py that every folder is correct; if a data set needs to be created, then make sure that the associated folder has been already created.

Then generate and precompute distance and kernel matrices

python red_button.py --generate --precompute

Data is also available here: delaunay, iam, kaggle-seizure.

Entire experimental campaign

Run the actual experiments

python red_button.py --delaunay
python red_button.py --iam
python red_button.py --kaggle
python red_button.py --supmat

fetch results

find $PWD/<prefix>*.zip > exp_list.txt
python red_button.py --readfrom exp_list.txt

Generate some of the figures

python red_button.py --figures

Single experiment

Each experiment has its own ID. To check the available ones

python run_simulation.py --help

For more details on what each ID corresponds to, please look at setting.py file.

Once selected the ID, say del_ratio, run the following command with the parameters of your choice

python run_simulation.py --numSimulations 100 --seed 20190512 --noJobs 15 --test mucpm --experiment del_ratio