Update documentation on nullable variable attributes

[mamscience]

Hi team,

Could you update the python-section of the documentation? I have no clue how to approach nullable variable attributes without an example . https://docs.tiledb.com/main/how-to/arrays/writing-arrays/nullable-attributes

Many thanks, Michel

[ihnorton]

Hi @mamscience, at the moment, we support pandas nullable datatypes transparently in TileDB-Py. We'll get the docs there updated ASAP, but in the meantime please have a look at these tests. We're also planning to add a new API for writing and querying nullable attributes soon (it will accept or return a numpy bool vector). If you have a particular package/API of interest, please let us know, we may not be able to integrate directly in the short term, but we'll keep it in mind to try to maximize interoperability.

[mamscience]

Thanks, apparently, changing the numpy type to float and passing "None" also did the trick.

schema = tiledb.ArraySchema(
        domain=dom, sparse=True, attrs=[
            tiledb.Attr(name="a, dtype=np.int32, nullable=True),
            tiledb.Attr(name="b", dtype=np.float32, nullable=True)
             ]
    )

snip

with tiledb.SparseArray(array_name, mode="w") as A:
        I, J = [1, 1, 1], [1, 2, 3]
        a = np.array([60, 65,64])
        b = np.array([120, 122, None])
        A[I, J] = {"val a":a, "val b":b}`

[mamscience]

The underlying issue btw is that I must pass all attributes when writing to array, while (my) sparse arrays are mostly empty. Solving this issue by relaxing the constraints (https://github.com/TileDB-Inc/TileDB/issues/1162#issue-425631917) was proposed before .

But let's say I need to use integer instead of float attribute, how should one approach this? Did you mean that I compile and populate a pandas df and convert it to tiledb array? Or is there another way to create tiledb array with pd datatypes

thanks in advance