Some experiments on deep neural networks and fault injections

Create whatever virtual environment and install requirements in requirements.txt.

If you're using virtualenv and pip, will go like this:

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

(venv) python manage.py [experiment class name]

e.g.

(venv) python manage.py ClipperVSRanger

This code is done using template method pattern. Each Experiment has some hooks to create models, fault injection configurations, etc. To add another experiment you have to extend ExperimentBase and implement the desired behaviour for it. You can add a directory to the root and register it in settings.py to add a new series of experiments.

A good example of implementing this class can be found as ClipperVSRanger experiment. You have to configure its dataset manually by changing the get_dataset method.

The default implementation for ExperimentBase logs the results in pickle format containing all samples classifications.

You can customize this behaviour by overriding the logging methods in your experiment.