I solved the OpenAI-Gym MountainCar-v0 environment using dqn(using low-state-dimensional input). When I used the same code for solving CartPole-v0 environment, the network got trained in the reverse direction (It sort of unlearned the environment and now performs worse than random actions). What could be the possible reason for this?

  • 2
    $\begingroup$ There are too many possible reasons to make an good answer here. RL is complex, and a mistake in any one part can result in code which runs but fails to achieve your goal of a learning agent. The specific result of "does worse than random" does not identify a cause. Most likely you have a bug or bad assumption somewhere in your code - it is not possible to say more without getting involved in your work in detail $\endgroup$ Sep 21 '19 at 7:44
  • $\begingroup$ When you're saying "it unlearned", do you mean that in addition to recycling the same code, you started with an initial neural network having the same weight as the one trained for MountainCar-v0? $\endgroup$
    – Jeanba
    Dec 16 '19 at 17:46
  • $\begingroup$ No, the initialization was new. By unlearn, what I mean is it performs worse than taking random actions. $\endgroup$ Jan 4 '20 at 8:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.