The purpose of this repo is twofold:
- To reproduce some of the results from the paper Neural Network Attributions: A Causal Perspective and repository.
- To further explore using average causal effect (ACE) to analyze neural networks, particularly in adversarial settings.
├── LICENSE
├── Makefile <- Makefile with commands like `make data` or `make train`
├── README.md <- The top-level README for developers using this project.
├── data (HIDDEN) <- Hidden from Git, but files are in a public Google Drive (see below)
│ ├── models <- Trained and serialized models, model predictions, or model summaries
│ ├── processed <- The final, canonical data sets for modeling.
│ ├── raw <- The original, immutable data dump.
│ ├── results <- Intermediate results files.
│ └── viz <- Images generated for visualization.
│
├── docker <- A Dockerfile and scripts for development in a container
│ ├── apt-requirements.txt <- Apt packages requirements file for building the docker image
│ └── requirements.txt <- The requirements file for reproducing the analysis environment
│ └── requirements-freeze.txt <- The detailed requirements file generated with `pip freeze > requirements-freeze.txt`
│
├── docs <- A default Sphinx project; see sphinx-doc.org for details
│
├── references <- Data dictionaries, manuals, and all other explanatory materials.
│
├── reports <- Generated analysis as HTML, PDF, LaTeX, etc.
│ └── figures <- Generated graphics and figures to be used in reporting
│
│
├── setup.py <- makes project pip installable (pip install -e .) so src can be imported
├── src <- Source code for use in this project.
│ ├── __init__.py <- Makes src a Python module
│ │
│ ├── attacks <- Scripts to generate attacks using ACE
│ │
│ ├── data <- Scripts to download or generate data
│ │
│ ├── model_analysis <- Scripts to analyze trained models
│ │
│ ├── models <- Scripts to train models and then use trained models to make predictions
│ │
│ ├── stages <- DVC stage files for defining reproducible experiments
│ │
│ ├── tests <- Scripts to test utilities and algorithms
│ │
│ ├── utils <- Scripts for everything else
│ │
│ └── visualization <- Scripts to create exploratory and results oriented visualizations
│
└── tox.ini <- tox file with settings for running tox; see tox.readthedocs.io
Project based on the cookiecutter data science project template. #cookiecutterdatascience
- TODO: Add manual installation instructions.For GPU acceleration:
- Use a Linux host for the Docker container.
- Nvidia Container Toolkit
Carefully consider where you will clone the repository since it will be bound as the container volume and so must be available to Docker to bind.
$ git clone https://github.com/jalane76/ace-net.git
$ export ACE_NET_HOME=/absolute/path/to/ace-net
$ cd ${ACE_NET_HOME}/docker
$ docker pull jalane76/ace-net
$ ./run-container
The run script assumes that the host has been set up with GPU support. Running CPU-only is as simple as editing run-container.sh to comment out the GPU support line and uncomment the no GPU support line.