Can we implement DQN model in tensorflow without env, like by just using X_train and y_train ad predict on X_test

Can we implement DQN model in tensorflow without env, like by just using X_train and y_train ad predict on X_test. So, I have seen across many areas about training DQN, but it do require environment class though, I want to know whether we can train DQN model on X_train, y_train unlike using env?

Hello @Arjun_Reddy,

Reinforcement Learning works this way:

image
(from google search results)

Because we do not have a training dataset beforehand, we need to “collect” data in real-time by taking actions and receving some feedbacks from the environment. The environment can be a simulator like the one you will see in the course’s final assignment, or the environment can be the real-world. However, simulator is always preferred when you don’t want to damage anything by wrong actions. If you need a simulator, then you are needing a environment.

Cheers,
Raymond

PS: moved your thread to the MLS course 3 week 3

The short answer is “no, we can’t”.

The Q values are learned from a training set.