Michael Panchenko

This will be solved as part of #933. Typed returns with annotations will make it easier to trace down the error

The symlinks are there for executing examples inside tests. I have the feeling that this mechanism should generally be adjusted, after which this issue may become obsolete. I'm putting it...

Episodes with varying lengths are an important feature, especially since gymnasium `step` can return `truncated=True`. There is no reason at all to expect this to be thrown at the same...

@ivanappliedai @xuzzo

There is a significant difference, it is a naming issue. For continuous envs, the output of forward is not a log-prob but the inputs of a distribution, typically mean and...

> REDQPolicy Class: Also here loc_scale maybe should be renamed? @MischaPanch > > https://github.com/thu-ml/tianshou/blob/4756ee80ff11cd8692aef3752f35c0af60a452e8/tianshou/policy/modelfree/redq.py#L147-L166 > What's wrong with the loc_scale? > So basically for all actors and all policies that...

> I just wanted to keep consistency. If i call it `action_dist_input` somewhere because is the input of the Torch distribution i should also do it anywhere else where thats...

> The output of line 85 that is named logits is also a mean and standard deviation, no? But is still called `logits` and not `loc_scale `. Maybe i am...

@Trinkle23897 I think this is the one my colleagues and I will address next. I tried to write a comprehensible description above, feel free to share your thoughts!

Hi @spacegoing, thank you for voicing your concern! To make it clear from the beginning: by no means I want to turn tianshou into another Rllib. I believe that it's...