Support pandas "Int64" type #607
Description
What is the feature or improvement you would like to see?
This produces an error:
import pandas as pd
import numpy as np
import igraph as ig
df = pd.DataFrame(np.array([[0, 1], [1, 1]]), columns=['a', 'b'])
df["a"] = df["a"].astype('Int64')
g = ig.Graph.DataFrame(df)
The error message:
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
Cell In [19], line 1
----> 1 g = ig.Graph.DataFrame(df)
File ~/.pyenv/versions/3.11.0/lib/python3.11/site-packages/igraph/io/objects.py:424, in _construct_graph_from_dataframe(cls, edges, directed, vertices, use_vids)
419 if use_vids:
420 if not (
421 str(edges.dtypes[0]).startswith("int")
422 and str(edges.dtypes[1]).startswith("int")
423 ):
--> 424 raise TypeError(
425 f"Source and target IDs must be 0-based integers, found types {edges.dtypes.tolist()[:2]}"
426 )
427 elif (edges.iloc[:, :2] < 0).any(axis=None):
428 raise ValueError("Source and target IDs must not be negative")
TypeError: Source and target IDs must be 0-based integers, found types [Int64Dtype(), dtype('int64')]
Int64 is a nullable pandas int data type. I believe the type check is too strict. As long as the Int64 does not contain any null value, the dataframe should be able to convert to an igraph object. The error message is also not precise for this case.
Use cases for the feature
Sometimes we just have to deal with a column with Int64 as the data type, it adds a bit of annoyance if we have to convert it to int64 every time.
References
https://pandas.pydata.org/docs/user_guide/integer_na.html