Scripts for training and evaluating neural networks used in the Wildfire DRL papers
The code is divided into separate files:
-
TrainNetwork.py: Trains the neural network controller
-
ReplayMemory.py: Represents the Q-learning replay memory
-
QNetwork.py: Contains all of the neural network code (Keras running on Theano), including an object for training Q-Networks and an object to read and evaluated trained network HDF5 files from Keras
-
Simulation.py: Represents the simulation environment
-
AircraftModel.py: Represents the aircraft state, dynamics, and observation model
-
Filters.py: Describes the EKF and particle filter approachs for filtering noisy camera observations
-
WildfireModels.py: Contains methods for a stochastic wildfire model
These files use python with the following dependencies:
- numpy
- math
- Theano
- Keras (using the Theano backend)