7 attack vs defense

.dockerignore

@@ -1,5 +1,4 @@
 venv
-output
-bin
-__pycache__
-.pytest_cache
+**/output
+**/__pycache__
+**/.pytest_cache

.github/workflows/python-app.yml

@@ -18,7 +18,7 @@ jobs:
     - name: Install dependencies
       run: |
         python -m pip install --upgrade pip
-        pip install flake8 pytest
+        pip install flake8 pytest pytest-cov
         if [ -f requirements.txt ]; then pip install -r requirements.txt; fi
     - name: Lint with flake8
       run: |
@@ -28,4 +28,4 @@ jobs:
         flake8 . --count --exit-zero --max-complexity=10 --max-line-length=127 --statistics
     - name: Test with pytest
       run: |
-        pytest
+        pytest --cov=src --cov-report=term-missing --capture=no

.gitignore

@@ -1,5 +1,9 @@
 venv
-output
+output*
+bns
+data
+plots
+*meta.txt
 bin
 __pycache__
 .pytest_cache

Dockerfile

@@ -1,8 +1,20 @@
-FROM python:bookworm
+FROM python:3.12-slim
 
+# Install system dependencies
+RUN apt-get update && apt-get install -y \
+    build-essential \
+    swig \ 
+    libglpk-dev \
+    python3-dev \ 
+    libcdd-dev \ 
+    libgmp-dev \
+    && rm -rf /var/lib/apt/lists/*
+
+# Set working directory
 WORKDIR /workspace
 COPY . .
 
+# Install python packages
 RUN pip install --upgrade pip
 RUN if [ -f requirements.txt ]; then pip install -r requirements.txt; fi
 

README.md

@@ -2,82 +2,112 @@
 
 Code for paper ["Towards Privacy-Aware Bayesian Networks: A Credal Approach"](https://doi.org/10.3233/FAIA251419) presented at [ECAI 2025](https://ecai2025.org/).
 
-## Set up Python environment
+## Preliminaries
 
-Create and activate a Python virtual environment with: 
+### Experiments
 
-```bash
-python3 -m venv venv
-source venv/bin/activate[.fish]  # use `.fish` suffix if using fish shell
-```
+`<name>` is the name of the experiment to run. Each `<name>` has its own directory, which is named the same way. Each of these contains the experiment logic, configuration file (`config.yaml`), eventually generated models and data, output directory, and a `Plot_results.ipynb` notebook to plot results.
+
+`<name>` can be one of the following:
+
+1. `cn_privacy`: run membership inference attack against a Bayesian network (BN), its related credal network (CN), and compute the theoretical privacy estimate of BN.
+
+2. `cn_vs_noisybn`: compare two privacy techniques, namely the CN and a noisy version of BN. All models are naive Bayes with target variable T. First, the CN and noisy BN hyperparameters are fine-tuned so that they achieve the same privacy level; then, their accuracy is computed in terms of most probable explanation (MPE) on variable T.
+
+For additional details, we refer to the paper.
+
+### Attacks and defenses
 
-Install all dependencies with: 
+Each experiment requires the user to specify one defense and one attack mechanisms, plus additional related hyperparameters. Below, the mechanisms and hyperparameters names are reported.
+
+Implemented defenses:
+- `def_idm`. Requires: `ess`.
+- `def_ran`. Requires: `delta`.
+
+Implemented attacks:
+- `atk_mle`. Requires: `n_bns`.
+- `atk_cen`.
+- `atk_ran`.
+- `atk_ent`.
+
+## Running code
+
+### Using Docker (recommended)
+
+The `compose.yaml` file contains a set of pre-set experiments. Additional ones can also be specified. The `generate_compose.py` file helps in generating them automatically.
+
+Generate models and data for all experiments (controlled by `config.yaml`):
 
 ```bash
-pip install -r requirements.txt
+python -m experiments.cn_privacy.generate
+python -m experiments.cn_vs_noisybn.generate
 ```
 
-Upgrade all Python packages with: 
+Run one or more experiments with:
 
 ```bash
-pip install --upgrade $(pip freeze | cut -d '=' -f 1)
-pip freeze > requirements.txt
+docker compose up [service name]
 ```
 
-This updates the requirements file with the upgraded packages.
+Results will be available under `experiments/<name>/output_*`.
 
-## Experiments
+To check the status, run one or more of the following:
 
-`<name>` is the name of the experiment to run. It can be one of the following.
-
-1. `cn_privacy`: run membership inference attack against a Bayesian network (BN), its related credal network (CN), and compute the theoretical privacy estimate of BN. The pipeline and results are described in the paper.
+```bash
+docker compose ps
+docker compose logs [service name]
+docker stats
+```
 
-2. `cn_vs_noisybn`: additional experiment, not reported in the paper. It compares two privacy techniques, namely the CN and a noisy version of BN. All models are naive Bayes with target variable T. First, the CN and noisy BN hyperparameters are fine-tuned so that they achieve the same privacy level; then, their accuracy is computed in terms of most probable explanation (MPE) on variable T.
+### Local computation
 
-## Run code
+Create and activate a Python virtual environment: 
 
-### With Docker (recommended)
+```bash
+python3 -m venv venv
+source venv/bin/activate[.fish]  # use `.fish` suffix if using fish shell
+```
 
-1. Build the Docker image:
+Install dependencies: 
 
 ```bash
-docker build . -t bnp:2025
+pip install -r requirements.txt
 ```
 
-2. Run the experiment:
+*Notice*: if some package is missing locally, see the `Dockerfile` for additional packages to be installed (names refer to Ubuntu/Debian).
+
+Upgrade dependencies: 
 
 ```bash
-docker run [-d] [--rm] -v bnp:/workspace bnp:2025 python -m experiments.<name>.main
+pip install --upgrade $(pip freeze | cut -d '=' -f 1)
+pip freeze > requirements.txt
 ```
 
-3. Results available at: 
+*Notice:* each of the following command will overwrite any related output. 
 
-`/var/lib/docker/volumes/bnp/_data/experiments/<name>/output/`.
+Generate models and data (controlled by `config.yaml`):
 
-### Without Docker
+```bash
+python -m experiments.<name>.generate
+```
 
-1. Run the experiment: 
+Run an experiment: 
 
 ```bash
-python -m experiments.<name>.main
+python -m experiments.<name>.exp def_mec=<def_name> [param=value] atk_mec=<def_name> [param=value]
 ```
 
-2. Results available at: 
+Results will be available under `experiments/<name>/output`.
 
-`experiments/<name>/output/`.
 
-## Test code
+## Testing code
 
-Run tests with:
+Run integration tests:
 
 ```bash
-pytest
+pytest [--cov=src] [--cov-report=term-missing] [--capture=no]
 ```
 
-Test results are available at: 
-
-`test/<name>/output/`.
-
 ## Formatting and linting
 
 Format code by running:
@@ -91,15 +121,11 @@ Lint code by running:
 
 ```bash
 flake8 . --count --select=E9,F63,F7,F82 --show-source --statistics --exclude=venv
-flake8 . --count --exit-zero --max-complexity=10 --max-line-length=127 --statistics --exclude=venv
+flake8 . --count --exit-zero --max-complexity=10 --ignore=E203 --max-line-length=140 --statistics --exclude=venv
 ```
 
 Analyze code by running:
 
 ```bash
 pylint $(git ls-files '*.py')
 ```
-
-## Plot results
-
-Use the `Plot_results.ipynb` notebook available for each experiment. Plots will be saved at: `experiments/<name>/output/plots`.

compose.yaml

@@ -0,0 +1,86 @@
+version: '3.9'
+services:
+  cn_vs_noisybn_def_idm_atk_ran_ess1:
+    build: .
+    command:
+    - python
+    - -m
+    - experiments.cn_vs_noisybn.exp
+    - def_mec=def_idm
+    - atk_mec=atk_ran
+    - ess=1
+    image: bnp:2025
+    volumes:
+    - ./experiments/cn_vs_noisybn/bns:/workspace/experiments/cn_vs_noisybn/bns
+    - ./experiments/cn_vs_noisybn/data:/workspace/experiments/cn_vs_noisybn/data
+    - ./experiments/cn_vs_noisybn/output_def_idm_atk_ran_ess1:/workspace/experiments/cn_vs_noisybn/output
+  cn_vs_noisybn_def_idm_atk_ran_ess100:
+    build: .
+    command:
+    - python
+    - -m
+    - experiments.cn_vs_noisybn.exp
+    - def_mec=def_idm
+    - atk_mec=atk_ran
+    - ess=100
+    image: bnp:2025
+    volumes:
+    - ./experiments/cn_vs_noisybn/bns:/workspace/experiments/cn_vs_noisybn/bns
+    - ./experiments/cn_vs_noisybn/data:/workspace/experiments/cn_vs_noisybn/data
+    - ./experiments/cn_vs_noisybn/output_def_idm_atk_ran_ess100:/workspace/experiments/cn_vs_noisybn/output
+  cn_vs_noisybn_def_idm_atk_ran_ess50:
+    build: .
+    command:
+    - python
+    - -m
+    - experiments.cn_vs_noisybn.exp
+    - def_mec=def_idm
+    - atk_mec=atk_ran
+    - ess=50
+    image: bnp:2025
+    volumes:
+    - ./experiments/cn_vs_noisybn/bns:/workspace/experiments/cn_vs_noisybn/bns
+    - ./experiments/cn_vs_noisybn/data:/workspace/experiments/cn_vs_noisybn/data
+    - ./experiments/cn_vs_noisybn/output_def_idm_atk_ran_ess50:/workspace/experiments/cn_vs_noisybn/output
+  cn_vs_noisybn_def_ran_atk_ran_delta0.001:
+    build: .
+    command:
+    - python
+    - -m
+    - experiments.cn_vs_noisybn.exp
+    - def_mec=def_ran
+    - atk_mec=atk_ran
+    - delta=0.001
+    image: bnp:2025
+    volumes:
+    - ./experiments/cn_vs_noisybn/bns:/workspace/experiments/cn_vs_noisybn/bns
+    - ./experiments/cn_vs_noisybn/data:/workspace/experiments/cn_vs_noisybn/data
+    - ./experiments/cn_vs_noisybn/output_def_ran_atk_ran_delta0.001:/workspace/experiments/cn_vs_noisybn/output
+  cn_vs_noisybn_def_ran_atk_ran_delta0.05:
+    build: .
+    command:
+    - python
+    - -m
+    - experiments.cn_vs_noisybn.exp
+    - def_mec=def_ran
+    - atk_mec=atk_ran
+    - delta=0.05
+    image: bnp:2025
+    volumes:
+    - ./experiments/cn_vs_noisybn/bns:/workspace/experiments/cn_vs_noisybn/bns
+    - ./experiments/cn_vs_noisybn/data:/workspace/experiments/cn_vs_noisybn/data
+    - ./experiments/cn_vs_noisybn/output_def_ran_atk_ran_delta0.05:/workspace/experiments/cn_vs_noisybn/output
+  cn_vs_noisybn_def_ran_atk_ran_delta0.1:
+    build: .
+    command:
+    - python
+    - -m
+    - experiments.cn_vs_noisybn.exp
+    - def_mec=def_ran
+    - atk_mec=atk_ran
+    - delta=0.1
+    image: bnp:2025
+    volumes:
+    - ./experiments/cn_vs_noisybn/bns:/workspace/experiments/cn_vs_noisybn/bns
+    - ./experiments/cn_vs_noisybn/data:/workspace/experiments/cn_vs_noisybn/data
+    - ./experiments/cn_vs_noisybn/output_def_ran_atk_ran_delta0.1:/workspace/experiments/cn_vs_noisybn/output