Kyurae Kim
Kyurae Kim
Okay let me pick this up.
Hi @devmotion I opened a PR, could you take a look when you have time?
Hi @devmotion , unfortunately, it seems that calls to `rand(q, d)` still result in the wrong type. I guess this is because `eltype(q)` returns `Float64`. Should I make a PR...
Hi @devmotion , could you comment on how to proceed?
Hi @yebai ! I think minibatching will require a more invasive solution than a syntax akin to `condition`. For instance, if the model doesn't have any latent variables/local variables like...
> We can treat the idx variable as a special argument, i.e., Oh yes, I think that would actually work for now if we assume subsampling over the prior is...
Yes how to do all this is pretty clear to me now! Just in terms of the details, though, I was thinking more like ```julia sub_iter = sample(subsample_range, B, replace=false)...
I think one downside would be that referring to the documentation becomes a little tricky since everything requires specialization. For instance, if I only care about `VI`, doing `?optimise` shows...
@mhauru I think that could be accommodated by pushing everything VI or ML/MAP specific to the object representing the algorithm like `vi( ... )`. So, the general interface for `optimise`...
Hi Dominique. I think here, the situation is slightly different since the unified interface is over ML and MAP. And the potential users of ML and MAP are most likely...