Force cast to pd.DataFrame on nonsensical manipulation of TableDataFrame #105
Description
Currently. .transpose() raises a warning but continues on its merry way, resulting in units that are nonsense. This shouldn't be allowed.
Should fall back on returning a vanilla pd.DataFrame instead.
Example:
from pdtable import Table
import pandas as pd
df = pd.DataFrame({"a":[1,2,3], "b":['q','w','e']})
t = Table(df, name="foo", units=["m", "text"])
print(t)
displays
**foo
all
a [m] b [text]
1 q
2 w
3 e
So far so good. Now let's .transpose() the backing TableDataFrame, and create a Table from that:
print(Table(t.df.transpose()))
We get a warning:
[...]\pdtable\frame.py:92: UserWarning: While combining pdTable metadata an unknown __finalize__ method "transpose" was encountered. Will try to propagate metadata with generic methods, but please check outcome of this and notify pdTable maintainers.
warnings.warn(
**foo
all
0 [text] 1 [text] 2 [text]
1 2 3
q w e
... but this still spits out a valid Table whose units are nonsense. This should not be allowed.
The proposed fix (i.e. force cast to pd.DataFrame) can be done by adding an elif clause as indicated below, which would eventually cover all methods that are nonsensical in the StarTable universe, such as transpose().
if method is None or method in frozenset({"reindex", "take", "copy"}):
# method: None - copy, slicing (pandas <1.1)
src = [other]
elif method == "merge":
src = [other.left, other.right]
elif method == "concat":
src = other.objs
elif method == "transpose": # <<<<<<<<<<<<<<<<<<<<
return None # <<<<<<<<<<<<<<<<<<<<
else:
# Unknown method - try to handle this as well as possible, but rather warn and drop units than break things.
src = [other]
warnings.warn(
f'While combining pdTable metadata an unknown __finalize__ method "{method}" was encountered. '
f"Will try to propagate metadata with generic methods, but please check outcome of this "
f"and notify pdTable maintainers."
)