ENH: Dataframe melt : the variable column should be typed as string

[simonaubertbd]

Feature Type

• Adding new functionality to pandas

• Changing existing functionality in pandas

• Removing existing functionality in pandas

Problem Description

When I use a melt in my script, I always have to retype manually the "variable" column to string.

Feature Description

A column name is always a string so it would make sense to cast it directly to the good dtype by default, here a string.

Alternative Solutions

The one I'm doing today : manually set the variable column to string

Additional Context

No response

[Liam3851]

It is not true that "a column name is always a string", so I don't think a cast to strings would really work...

In [11]: df = pd.DataFrame([[1, 2], [3, 4]], columns=pd.Index([0, 1], name=5), index=pd.Index([2, 3], name=6))

In [12]: df.melt(id_vars=0, value_vars=1, var_name=8, value_name=9)
Out[12]:
   0  8  9
0  1  1  2
1  3  1  4

In [13]: df.melt(id_vars=0, value_vars=1, var_name=8, value_name=9).dtypes
Out[13]:
0    int64
8    int64
9    int64
dtype: object

Often of course they will be string, but in general index and column values in pandas need only be hashable, not strings.

That said, perhaps there would be an argument for downcasting to the new string dtype instead of to object, when the columns are all strings, in the same way that above the dtype is downcast to int when the columns are all integers.

[rhshadrach]

Thanks for the report. In the future, please provide a reproducible example rather than a screenshot. The requested behavior is already implemented by setting future.infer_string to True. This will be the default in pandas 3.0.

pd.set_option("future.infer_string", True)
df = pd.DataFrame([[1, 2], [3, 4]], columns=["a", "b"], index=pd.Index([2, 3], name=6))
print(df.melt(id_vars="a", value_vars="b", var_name=8, value_name=9).dtypes)
# a    int64
# 8      str
# 9    int64
# dtype: object

Closing.