RL for Pokémon Battling

Experiment1.py - investigating the effects of changing policy on RL agent effectiveness

• Simple RL agent implementation with a state embedder and reward computation
• Tested 3 policies: Greedy Q Policy, Epsilon Greedy Q Policy, Boltzmann Q Policy

Experiment2.py - policy Hyperparameter Tuning

Experiment3.py - Round Robin for the optimal agent based on the player it was trained against

Experiment4.py - Round Robin with Minimax Algorithm

Experiment5.py - (in progress) PPO experiment

Final results and report from experiments can be found here

Pokemon showdown environment forked from https://github.com/smogon/pokemon-showdown