Deep RL algorithms require fine tuning, which is expensive on real apps. Therefore, we developed FATE, a simulation environment for fast Android testing. FATE models only the navigation constraints of Android apps, so it can be used to efficiently compare alternative testing algorithms and to quickly tune their corresponding hyperparameters.
- Ptolemy: https://ptolemy.berkeley.edu
- python 3.5 to 3.7 (tensorflow limitation)
In the folder ptolemy
you can find an example of app. Following the pattern of bank.xml you can define your own app.
Once finished you can launch the script : python3 ptolemy_to_fate.py -n 'path_to_xml'
.
As output, you will have a python file named as your original xml file. Please check the global_vars
inside the
python file.
Once done use you modelled application in train_TD3.py, train_SAC.py, train_DDPG.py
.