IMC23 Paper Artifacts

The paper is associated to the following types of artifacts:

• Data: This includes

  • The datasets curation and splits for ucdavis-icdm19, mirage19, mirage22 and utmobilenet21. Please refer to the datasets webpage and related pages for more details.
  • All models and logs generated through our modeling campaigns.

• Code: This includes

  • A collection of Jupyter notebooks used for the tables and figures of the paper.
  • A collection of data to support pytest unittest related to the results collected for the paper.

Figshare material

The artifacts are stored in a figshare collection with the following items:

• curated_datasets_ucdavis-icdm19.tgz: A curated version of the dataset presented by Rezaei et al. in "How to Achieve High Classification Accuracy with Just a Few Labels: A Semi-supervised Approach Using Sampled Packets".

• curated_datasets_utmobilenet21.tgz: A curated version of the dataset presented by Heng et al. in "UTMobileNetTraffic2021: A Labeled Public Network Traffic Dataset".

• imc23_ml_artifacts.tgz: Models and output logs generated via tcbench.

• imc23_notebooks.tgz: A collection of jupyter notebooks for recreating tables and figures from the paper.

• imc23_pytest_resources.tgz: A collection of reference resources for pytest unit testing (to verify model training replicability).

• ucdavis-icdm19-git-repo-forked.tgz: A fork of the repository https://github.com/shrezaei/Semi-supervised-Learning-QUIC- to verify results of "How to Achieve High Classification Accuracy with Just a Few Labels: A Semi-supervised Approach Using Sampled Packets" https://doi.org/10.48550/arXiv.1812.09761

Downloading artifacts

Each artifact can be manually downloaded from the figshare collection. However, make sure to refer to the latest version of an archive when downloading manually.

tcbench offers automated procedures to fetch the right content from figshare:

• For datasets please refer to datasets page page, the specific page for each datasets and the import command.

• For the remaning, you can use the fetch-artifacts subcommand with the following process

• First of all, prepare a python virtual environment, for example via conda

  conda create -n tcbench python=3.10 pip
  conda activate tcbench
  

• Clone the tcbench repo and use the imc23 branch

  git clone https://github.com/tcbenchstack/tcbench.git tcbench.git
  cd tcbench.git
  git checkout imc23
  

• Install tcbench

  python -m pip install .[dev]
  

• Fetch the artifacts

  tcbench fetch-artifacts
  

This will install locally

• The notebooks for replicating tables and figures of the paper under /notebooks/imc23. The cloned repository already contains the notebooks but since the code might change, the version fetched from figshare is identical to what used for the submission.

• The ml-artifacts under /notebooks/imc23/campaigns.

• The pytest resources for enabling unit tests.

Packages depencency version and /imc23 branch

When installing tcbench via pypi of from the main branch of the repository, only a few sensible packages have a pinned version.

If you are trying to replicate the results of the paper, please refer to the /imc23 branch which also contains a requirements-imc23.txt generated via pip freeze from the environment used for collecting results.

Based on our experience, the most probable cause of results inconsistency is due to package version.