DataFrame with namedtuples uses ._field as column names

[mroeschke]

Description

Allow namedtuple's _field attribute to be mapped to DataFrame column labels like pandas

closes #13823

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[galipremsagar]

Suggested change

	elif (
	len(data) > 0
	and columns is None
	and isinstance(data[0], tuple)
	and hasattr(data[0], "_fields")
	):
	# pandas behavior is to use the fields from the first tuple
	# as the column names
	columns = data[0]._fields
	values = itertools.zip_longest(*data)
	data = dict(zip(columns, values))
	self._init_from_dict_like(
	data, index=index, nan_as_null=nan_as_null
	)
	if (
	len(data) > 0
	and columns is None
	and isinstance(data[0], tuple)
	and hasattr(data[0], "_fields")
	):
	# pandas behavior is to use the fields from the first tuple
	# as the column names
	columns = data[0]._fields

Will just doing this work? i.e., I guess the code below would be handling the data construction no? I might be missing something but from the issue I thought we were only missing out on the column names.

[mroeschke]

I was curious about this too; since this code path just assigns self.columns = columns I wasn't sure if it was okay if columns was a tuple or needed to be a different type

[mroeschke]

Okay this looks to work and produces a column Index too so I changed it to this.

[wence-]

I think this doesn't mimic pandas behaviour exactly, and has some edge cases where it silently drops data.

[wence-]

/merge