conda install keras tensorflow-gpu
python train.py
Modify the train.py to choose different models; or modify the models.py to implement new models.
python predict.py
Modify the domain.json to add/remove the domains to be classified; or modify the predict.py to choose different models or weighs.
python evaluate_on_other_dga.py
Modify the evaluate_on_other_dga.py to choose different DGA set to evaluate.
Training on the dataset built by https://github.com/andrewaeva/DGA, including 1,000,000 legit domains and 801,667 DGA generated domains. While calling dataset.load_data, if specify filter=True then legal domains that end with different suffixes than DGA generated domains are not loaded, and also suffixes of all loaded domains are removed.
Evaluating on the dataset provided by http://data.netlab.360.com/feeds/dga/dga.txt, https://github.com/philarkwright/DGA-Detection, https://github.com/nickwallen/botnet-dga-classifier and https://github.com/ClickSecurity/data_hacking .