lantunes / mountain-car-continuous Goto Github PK

Hey Thanks for sharing out the repo.

Everything makes sense to me except for sigma not being tied to input (states).

Does this suggests that regardless of the state, the environment itself has an ideal sigma (ultimately std), and we want to optimize using loss? Wondering why this is better than using states as input? Thanks for any thoughts, etc...

First, thanks for this nice work!

But is the plot in the documentation obtained with exactly the published code?
Trying to reproduce the training seems much slower... I am not sure if it is succesful at all, or just learning not to move.