AnnData2SCE gets `adata$X` to memory

[Artur-man]

I have quickly checked if how AnnData2SCE realizes the table in zarr format.

This function here, checks the format and decides if X should go to memory or HDF5Array.
https://github.com/theislab/zellkonverter/blob/41861c688218c23a44da20ee7d97cddc7a1985b0/R/AnnData2SCE.R#L664C1-L668C2

.is_anndata_matrix <- function(mat) {
    is(mat, "h5py._hl.dataset.Dataset") ||
        is(mat, "anndata._core.sparse_dataset.SparseDataset") ||
        is(mat, "anndata._core.sparse_dataset.BaseCompressedSparseDataset")
}

if you check the class of the datasets in python you get:

Browse[1]> adata$X
<Compressed Sparse Row sparse matrix of dtype 'float64'
with 30 stored elements and shape (10, 3)>

which is not what .is_anndata_matrix looking for.

I would think getting adata$X into memory and then save under assays in SCE as DelayedArray is not ideal, but too many things may need to change in SCE, zellkonverter to get immediate definition of a DelayedArray object ?

[Artur-man]

It seems the current spatialdata.read_zarr implementation, and hence anndata.io.read_zarr doesnt have lazy reading either.
#data-structures > disk-backed read for spatialdata/anndata.zarr @ 💬

Only through dask but still experimental ...