Which datatypes does `to_dense` accept?

[jan-engelmann]

Thank you for this package!

I was wondering if you could extend either the functionality or the documentation of supported input datatypes of to_dense

For example the following common use-case works:

In [1]: import pandas as pd
In [2]: import numpy as np
In [3]: from fast_array_utils.conv import to_dense
In [4]: to_dense(pd.DataFrame(np.random.randn(100,5)))

but it is not documented.

Furthermore, I was wondering why you split the functionality into: fast_array_utils.conv.scipy.to_dense and fast_array_utils.conv.to_dense

For me the advantage of this function is precisely not having to think about the which kind of array we are putting in.

Thank you!

[jan-engelmann]

Ok the usage section in the README contains a more exhaustive list than the docs

Then it would be great if you can also reliably support pandas. potentially with a warning in case of mixed datatypes

[flying-sheep]

The docs have everything! E.g. the first overload is

clicking e.g. CpuArray leads to

fast_array_utils.typing.CpuArray: TypeAlias = numpy.typing.NDArray[typing.Any]
    | scipy.sparse.csc_matrix | scipy.sparse.csr_matrix
    | scipy.sparse.csc_array | scipy.sparse.csr_array

so this should be all the information you need

Then it would be great if you can also reliably support pandas. potentially with a warning in case of mixed datatypes

that makes sense!

[jan-engelmann]

ah makes sense! thank you

[jan-engelmann]

Then if I may add to the wishlist:

• torch tensor
• jax array

Actually einops has a very similar function asnumpy