Deep Reinforcement Learning Coursework

All the code was developped using Google Colab. The Readme include links for all notebooks.

Basic: TicTacToe Agent

Notebook

Advanced: Pong x Breakout

DQN Implementation and Improvements

The two improvements implemented are Double DQN and Dueling DQN, the main notebook contain the implementation of both improvements. The other 3 notebooks are copy of the main one but modified to include only one or zero improvement for ablation studies.

Main notebooks

Pong_DuelingDDQN
Breakout_DuelingDQN

Ablation study notebooks

Pong_VanillaDQN
Pong_DDQN
Pong_DuelingDQN

RLLIB

As said in the report, I had many bugs with RLLIB for Atari environments whose states are frame after succeding in the first time.

Breakout
Breakout_Deterministic

Extra Task: PPO for CartPole-v0

PPO
Resource