Ariel Kwiatkowski

I think I found the issue - the custom RNG class inherits from the numpy Generator for compatibility purposes, but when it gets pickled/unpickled, it defaults to the numpy behavior...

@carlosluis >Regarding the "mean reward below threshold" problem, after some further investigations the root cause is the change in seeding behaviour in gym. Can you elaborate on how you got...

The difference is probably that the underlying RNG mechanism changed somewhere between releases, so the effect of setting an old random seed won't be the same. I'd be super uncomfy...

@benblack769 The issue was e.g. at https://github.com/openai/gym/blob/master/gym/utils/env_checker.py#L72 But as @pseudo-rnd-thoughts says, it's currently being fixed. Re 2: `self.seed(...)` is deprecated so depending on that is not the greatest idea. Nothing...

I don't see anything weird/concerning in the changes since my last pass, so it *should* be alright

Can you describe what exactly is the difference between this API and the current parallel API? Also, what exactly is the point of `Container` as opposed to `Dict`? My understanding...

What's the status of this? I'm annoyed by the random ordering every time I go back to rustlings, I'd be happy to contribute a PR if that'd be helpful. I...

I don't really see how it would cause confusion. I'm thinking about a simple naming scheme: ``` 01_functions 02_if 03_enums 04_conversions ... ``` (the order is probably not correct because...

You can do headless rendering with pyvirtualdisplay, integrating it in MLA would be nice, but it's not too hard to make it work yourself

I second this very strongly, sometimes I had to do very weird workaround just to get some simple debug information, e.g. the total reward that the agent has obtained in...