I have yet to run the agent for a long enough time to see if the RL is working. When the agent had rewards for building units, and killing the enemy, the agent would choose to do no action (the agent found that doing nothing gave the highest rewards).
After switching to sparse rewards and running a few times, I noticed the agent was close to defeating the easy A.I. It may be better to have the agent practice against the very easy A.I. to develop a Q-table without any hindrances, and then switching to a more difficult A.I. when it has learned how to defeat the enemy quickly.
I have yet to run the agent for a long enough time to see if the RL is working. When the agent had rewards for building units, and killing the enemy, the agent would choose to do no action (the agent found that doing nothing gave the highest rewards).
After switching to sparse rewards and running a few times, I noticed the agent was close to defeating the easy A.I. It may be better to have the agent practice against the very easy A.I. to develop a Q-table without any hindrances, and then switching to a more difficult A.I. when it has learned how to defeat the enemy quickly.