How does reinforcement learning work in a pong game?

How does reinforcement learning work in a pong game?

The agent learns which actions maximize the reward, given what it learned from the environment. More precisely, in our Pong case: The agent is the Pong AI model we’re training. The action is the output of our model: tells if the paddle should go up or down. The environment is everything that determines the state of the game.

What does the agent do in a pong game?

The agent is the Pong AI model we’re training. The action is the output of our model: tells if the paddle should go up or down. The environment is everything that determines the state of the game. The observation is what the agent sees from the environment (here, the frames).

Is there a deep RL program for Pong?

In this post, we’ll dive into Deep RL ourselves by coding a simple Vanilla Policy Gradient model that plays the beloved early 1970s classic video game Pong. And, truth be told, our trained model is pretty darn good at Pong. Take a look:

Is there an AI that can solve Pong?

Yes, I know someone already wrote a 130-lines-of-Python algorithm (with very limited dependencies) that solves Pong. But, that was back in 2016, and it took him 3 whole nights on his MacBook to train the model. In 2018, we can use Keras, along with cutting edge GPUs on FloydHub, to train our Pong AI effortlessly. Let’s get started. Psst.

Is there a class of actions in Pong?

This can make assigning credit to a particular action pretty challenging. You can probably tell that there is class of actions in Pong where the Agent is simply keeping up the rally, versus a class of actions which is likely to win the game, which comprises the actions immediately prior to winning a game.

How is reinforcement learning used in machine learning?

The automated approach described above can be more formally described as Reinforcement Learning, since we have an Agent that is executing actions on its environment, observing the outcomes, collecting rewards, and using knowledge of the rewards collected to modify any future actions, in order to perform better.