index_constructors/columns_constructors not exposed on sf.Frame.from_pandas for multi-indices.

[chaburkland]

[flexatone]

Thanks for posting this issue.

StaticFrame interfaces accept either index_constructor/column_constructor or index_constructors/column_constructors, but not both. If *_constructor arguments are available, it is because it is not obvious that an IndexHierarchy is desired, and if so, an IndexHierarchy constructor can be partialed with index_constructors as needed (as shown below). If *_constructors arguments are available, it is because an IndexHierarchy is assumed, or a *_depth arguments are available.

While this heterogeneity is slightly undesirable, it does not limit utility (save sometimes requiring partialing).

>>> f = sf.Frame.from_element(0, index=(1,2), columns=sf.IndexHierarchy.from_product(('a', 'b'), ('2020-01-01', '2021-02-02'), index_constructors=(sf.Index, sf.IndexDate)))
>>> df = f.to_pandas()
>>> from functools import partial
>>> sf.Frame.from_pandas(df, columns_constructor=partial(sf.IndexHierarchy.from_labels, index_constructors=(sf.Index, sf.IndexNanosecond)))
<Frame>
<IndexHierarchy> a                    a                    b                    b                    <<U1>
                 2020-01-01T00:00:... 2021-02-02T00:00:... 2020-01-01T00:00:... 2021-02-02T00:00:... <datetime64[ns]>
<Index>
1                0                    0                    0                    0
2                0                    0                    0                    0
<int64>          <int64>              <int64>              <int64>              <int64>

[chaburkland]

I see. Personally I really hope that we can change this, as I balk at the idea of having to partial in a IndexHierarchy constructor.

[flexatone]

How do you suggest we change this?