Make obs and var optional, make obs_names and var_names required

[rcannood]

At the moment, we require that obs and var are passed to the InMemoryAnnData and HDF5AnnData, while other slots are optional.

However, looking at what happens when I create an empty AnnData in Python, it's actually the obs_names and the var_names that are required, and not obs and var -- I guess.

import anndata as ad
adata = ad.AnnData(shape=[10, 20])
adata.write_h5ad("empty.h5ad")

file <- rhdf5::H5Fopen("empty.h5ad")
rhdf5::h5ls(file)

  group   name       otype dclass dim
0     / layers   H5I_GROUP           
1     /    obs   H5I_GROUP           
2  /obs _index H5I_DATASET STRING  10
3     /   obsm   H5I_GROUP           
4     /   obsp   H5I_GROUP           
5     /    uns   H5I_GROUP           
6     /    var   H5I_GROUP           
7  /var _index H5I_DATASET STRING  20
8     /   varm   H5I_GROUP           
9     /   varp   H5I_GROUP

As such; I propose altering the code of InMemoryAnnData and HDF5AnnData require the obs_names and the var_names to be passed, not the obs and the var.

[lazappi]

I think this makes sense but is it much different to just doing obs = data.frame(row.names=obs_names).

What does Python anndata put in obs_names/var_names when you just set the shape?

[rcannood]

I agree this is very practical, but given #73 we can't add the obs_names to the obs as rownames or to the X as rownames because the obs_names might be integers instead of characters.

[rcannood]

What does Python anndata put in obs_names/var_names when you just set the shape?

This is what's shown in the original message, no?

[lazappi]

What does Python anndata put in obs_names/var_names when you just set the shape?

This is what's shown in the original message, no?

Not entirely, but I could just check myself rather than asking you silly questions 😸

[rcannood]

:p

Closed with #81