MsPacManEngine-RL

Implementation of reinforcement learning techniques for behavioral learning in the game MsPacMan Vs. Ghosts.

MsPacManQLearn

Implementation of Q-Learning for the MsPacMan behavior learning.

Train

• TrainChaseModel: class that trains the chase model.
  • Params: model name, number of episodes.
• TrainFleeModel: class that trains the flee model.
  • Params: model name, number of episodes.
• TrainOriginalModel: class that trains the original model.
  • Params: model name, number of episodes.
• TrainPillsModel: class that trains the pills model.
  • Params: model name, number of episodes.
• LoadAndTrainModel: class that loads a model that has been trained and trains it more.
  • Params: model name of the exisiting model, number of episodes.

Run model

• MainQ: class that trains a model and runs it.
• RunFsmModel: class that runs the game with a fsm.
• RunRandomModel: class that loads an original model and runs it in a randomly intialized game.
  • Params: model name.

Evaluation

• Evaluate: class that evalutes the model following the settings of config.properties.

MsPacManDeepRL

Implementation of Deep Reinforcement Learning for the MsPacMan behavior learning.
In every train, evaluate or run game you have to run both python server and java server.

Train

Java

• MainTrain: class that trains the general model.
  • Params: port.
• MainTrainEdible: class that trains the edible model(default:long state).
  • Params: port.
• MainTrainNotEdible: class that trains the not edible model.
  • Params: port.

Python

• trainerGeneral: class that trains the general model.
  • Params: number of episodes, port, name of the model, number of neurons of the hidden layer.
• trainerShortState: class that trains the edible model with short state.
  • Params: number of episodes, port, name of the model, number of neurons of the hidden layer.
• trainerLongState: class that trains the not edible model and the edible model with long state.
  • Params: number of episodes, port, name of the model, number of neurons of the hidden layer.
• trainerGPU: class that trains the general model using CUDA.
  • Params: number of episodes, port, name of the model, number of neurons of the hidden layer.

Run model

Java

• MainExecute: class that executes a general game.
• MainFSM: class that executes an fsm game. It requieres to load the edible model and the not edible model with the python launchers.

Python

• LoadMsPacManGeneral: class that loads and runs the general model.
  • Params: model file's name, port.
• LoadMsPacManShortState: class that loads and runs the edible model with short state.
  • Params: model file's name, port.
• LoadMsPacManLongState: class that loads and runs the not edible model and the edible model with long state.
  • Params: model file's name, port.

Evaluation

Java

• Evaluate: class that evalutes the model following the settings of config.properties.

Python

• LoadMsPacManEvaluate: class that evaluates the general model.
  • Params: model file's name, port.
• LoadMsPacManEvaluateFSM: class that evaluates the fsm. It has to be run two times. One for an edible model with long state and one for a not edible model.
  • Params: model file's name, number of different type of ghosts, number of episodes, port.