Astraea Quinn S

I’d rather have a separate implementation that includes entropy but it’s up to you. Yes I will add an example in the docs. > I don't have anything to add...

Fair point! I have a couple of weeks before grad classes pick up the pace so hopefully I will have everything tied up before then.

@araffin I am considering the following changes: 1. Use a component called say "batch generator" that given the indices produces the batch. 2. Instead of subclassing the buffer class for...

> How is that different from `_get_samples()` ? `_get_samples()` is bound to the class of the buffer. I would like to use composition instead of sub-classing the buffer to create...

I can make a large one that deals with code optimisation/quirks and leave #110 for algorithm performance, it's up to you :)

After some more investigation, I believe that we don't need to copy as the assignment is done on the original data, i.e. where the buffer is located, so ``` x[indice]...

> But if everything is copied, then perfect. > I played a bit with it in a local implementation and there seemed to be some performance degradation with respect to...

Using pytorch tensors doesn't necessarily improve performance mainly due to the need to perform the computations @araffin mentioned, moreover, going cpu->gpu->cpu will probably degrade performance due to latency transfers. The...

This approach sounds better than what I initially came up with, seems to have fewer moving parts and will be easier to reason about. I will get on it once...

How would you like this to be implemented? As a wrapper around the buffer, as a derived class from the buffer, or as it's own object that adheres to the...