The scripts in this directory can be used to evaluate the performance of different data representations (eg. direction, time, directional time) with the DeepFingerprinting model.

These experiments take WANG14 structured plain-text data files.

The main runnable scripts are cw_attack.py and ow_attack for closed and open-world respectively. The closed-world attack script evaluates closed world performance across several cross-validation folds. The open-world attack script does not perform cross-validation, however examines performance when different thresholding values are used.

• Python 3.X
  • Install required modules from the requirements.txt
  • Usage of python virtual environments is recommended
• CUDA 10.2 & CuDNN
• Undefended DF dataset
  • Download from: gdrive

Ex1. Setup virtual environment:

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Ex2. View argument help information:

python cw_attack.py -h

Ex3. Execute closed-world attack using direction representation:

python cw_attack.py -t /data/undefended/ -o df.h5 -a 0

Ex4. Execute open-world attack using time-only representation:

python ow_attack.py -m /data/undefended/ -u /data/undefended_ow/ -o df.h5 -a 2