Creating a meta-samplers repository

[AdrienCorenflos]

Hi,

I've been thinking recently about splitting responsibilities between sampling algorithms and meta-algorithms. Chief of these being SMC. It's pretty clear that they cannot be used by themselves, with no inner kernel feeding them.

It has some benefits:

• decluttering the API
• signaling responsibilities more clearly
• making contributions more localised
• being able to create more specialised ecosystems while maintaining coherence

And some inconvenients

• Split download and import (although that can be solved relatively easily via deps)
• Enforcing APIs compatibility would not be immediate
• Creating upstream monitoring strain for maintenance

What do you think?
Adrien

[junpenglao]

First thing comes to mind is: Is our SMC API slim enough that user could potentially use it without understanding too deep about the inner kernel implementation?