GP only takes AbstractVector as an argument about abstractgps.jl HOT 6 CLOSED

I really can't see what the difference is from the user's perspective

In one case, obsdim is pretty obviously the observation dimension, on the other side with ColVecs I will need to go through the docs, understand what it is and what it does and how to correctly use it.

In any case, I'll concede this point if you're really keen on it being a thing, and given that it's consistent with KF

Thanks! I will take care of making the PR

On the other hand, I have to say the structure is already pretty nice to use and I am looking forward to integrate it to AGP.jl

from abstractgps.jl.

Yup, requires either a ColVecs or RowVecs otherwise how do you know what convention the user is employing?

from abstractgps.jl.

Shouldn't we do like in KF.jl and put a wrapper for matrices by adding a obsdim keyword? Or decide of an arbitrary convention?

from abstractgps.jl.

I really don't see what the advantage is over just using ColVecs / RowVecs. One way of doing things is strictly better than two unless there's a really good reason to have two ways of doing things.

from abstractgps.jl.

This is probably going to be the same dialog again but :
I really don't believe we should force users to "convert" their data in a different format. I agree that it should be used in the internals but the user should not have to see it.
Again as a lambda user if I want to use a package I would just like to throw my matrix to the model eventually giving what is the dimensionality without having to wrap it first into some format.

from abstractgps.jl.

Agreed, this is going to go down the same line again.

Again as a lambda user if I want to use a package I would just like to throw my matrix to the model eventually giving what is the dimensionality without having to wrap it first into some format.

At the risk of labouring the point, wrapping your matrix is telling the package how to treat the dimensions of your data, nothing more. I really can't see what the difference is from the user's perspective between writing

f(ColVecs(X))

and

f(X; obsdim=2)

, particularly if we provide a good error message for f(X) that says how to handle multi-input data.

In any case, I'll concede this point if you're really keen on it being a thing, and given that it's consistent with KF

from abstractgps.jl.