20240212 Run a Jupyter notebook to launch the Unity Banana Collecting game (RL-DDQN)
The banana collector (aka. a Unity ML-Agents brain) in the game has been trained in Google Colab with Tesla T4 GPU for 2000 episodes (around 30 mins). It utilizes a custom Double Deep Q-Network (DDQN) as its core intelligent algorithm. Both DQNs are identical, having 2 linear layers (64, 64). The input is the state (length = 37) and output action (length = 1, 0~3 = go forward, backward, left, right). It is supposed to pick up the good bananas and leave out the bad ones!
👉 the train notebook - https://github.com/Nov05/Google-Colab...
👉 the eval notebook - https://github.com/Nov05/udacity-deep...
#ReinforcementLearning #MachineLearning #DQN #Unity #Colab