Go! Trail Ranger

Project Structure

1. unity-environment

Links to the the game specific data and assets in Unity Game Engine.

2. Reinforcement Learning

AIBot.py - Entry point where the AI bot learns the game using rewards and finally plays the game.
Agent.py - Agent structure.
CNN.py - Neural Networks brain that trains on the Game frames.
ExperienceReplay.py - Experience Replay dealing with Reward training.
MovingAverage.py - To calculate Moving average for the rewards.
ReplayMemory.py - Stores the samples of the Replay memory.
SoftmaxBody.py - Softmax function for the CNN classifier.
Utils.py - Other Util functions.

3. Imitation Learning

Imitation_Learning_Train.ipynb - Train the model and generate the weights.
Classify_and_Play.ipynb - Use the weights to classify the game frames and play.

4. Zero-shot Learning

Zeroshot_Learning_Train.ipynb - Train the model and generate the weights.
Classify_and_Play.ipynb - Use the weights to classify the game frames and play.

5. Transfer Learning

Transfer_Learning_Train_Pass1.ipynb - Train the model using Subway Surfers game data and generate the initial weights.
Transfer_Learning_Train_Pass2.ipynb - Train the model using Go! Trail Ranger game data and generate the final weights.
Classify_and_Play.ipynb - Use the weights to classify the game frames and play.

6. Transfer Learning Merge Network

Parallel_Networks_Train.ipynb - Train the model and generate the weights.
Classify_and_Play.ipynb - Use the weights to classify the game frames and play.

7. Results

Graphs.ipynb - Graphs.
Metrics.ipynb - Metrics.