1
$\begingroup$

I'm looking at some baseline implementations of RL agents on the Pendulum environment. My guess was to use a relatively small neural net (~100 parameters).

I'm comparing my solution with some baselines, e.g. the top entry on the Pendulum leaderboard. The models for these solutions are typically huge, i.e. ~120k parameters. What's more, they use very large replay buffers as well, like ~1M transitions. Such model sizes seem warranted for Atari-like environments, but for something as small as the Pendulum, this seems like complete overkill to me.

Are there examples of agents that use a more modest number of parameters on Pendulum (or similar environments)?

$\endgroup$
1
$\begingroup$

Actually, I just started inspecting the entries further down in the leaderboard list, and there are in fact more modest architectures, e.g. this one, which uses 3 hidden layers with 8 units each.

$\endgroup$

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.