uns field is replicated for all observation which makes it annotation of observation. about anndata HOT 4 CLOSED

what do you mean specifically and what do you suggest? can you give an example?

from anndata.

So, if I make a new anndata object using:
adata = sc.AnnData(X=adata_training.X, obs = adata_training.obs, uns = adata_training.uns)
The uns field is supposed to be one unstructured annotation for all observations. However, we can still use adata[0].uns , adata[1].uns ....etc. This would be fine if what we added to the uns field is a small dictionary, because replicating it to all observations will not significantly increase the size or the query time of the andata object.
However, once we have more meta data to add in the uns field, the size of the file as well as query/read/write time increase massively.

It could be, that I missed the point of using uns field but then the issue is still valid but related to the documentation of the Anndata object more than to the object itself.

from anndata.

Usually you wouldn't split up your AnnData's into many tiny objects. Also, usually, .uns doesn't contain a lot of data, just some very light-weight annotations.

from anndata.

Hi @falexwolf .

Usually you wouldn't split up your AnnData's into many tiny objects.

Well, sometimes you just need to parse and process the observations one by one.

Also, usually, .uns doesn't contain a lot of data, just some very light-weight annotations.

That was not my use case.

Anyway thanks for the great package and for your clarification :)

from anndata.