df_to_anndata does not assign binary numerical columns correctly in .uns

[Imipenem]

Describe the bug

When calling ep.df_to_anndata(...) the binary columns (columns with only two values like 1,0) do not get correctly assigned to "numerical_columns" in uns.

To Reproduce

Example:

df = pd.DataFrame({"Col1":[i for i in range(4)], "Col2":["str" + str(i) for i in range(4)], "Col3":[1.0,0.0,np.nan,1.0]})
adata = ep.ad.df_to_anndata(df)
print(adata.uns)

# OverloadedDict, wrapping:
	# OrderedDict([('numerical_columns', ['Col1']), ('non_numerical_columns', ['Col3', 'Col2'])])
# With overloaded keys:
	# ['neighbors'].

Col3 should be numerical!

Expected behavior

The binary columns with numerical values such as only 0 and 1 should be assigned to numerical columns.
The bug is caused by the following line:

ehrapy/ehrapy/anndata/anndata_ext.py

Line 71 in e1515b0

uns["numerical_columns"] = list(set(numerical_columns) ^ set(binary_columns))

The XOR excludes the binary columns since they are in both sets. This should be fixed.

System [please complete the following information]:

• Ehrapy version: 0.2.0

[Zethson]

We are unfortunately still running into this. We can supply an object to test this out and fix this very soon.