get_dummies（）之后的Pandas 0.21 reindex（）

Question

Background

My project is doing a Pandas upgrade from 0.19.2 to 0.21.0. In the project, I have a DataFrame with one categorical column. And I use get_dummies() to encode it, and then use reindex() to filter columns. However, if the columns arg in reindex() contain non-encoded column, the reindex() fails.

Sample Code

The code below works for 0.19.2 but fails under 0.21.0.

df = pd.DataFrame.from_items([('GDP', [1, 2]),('Nation', ['AB', 'CD'])])
df = pd.get_dummies(df, columns=['Nation'], sparse=True)  # SparseDataFrame
df.reindex(columns=['GDP'])  # Fails :/

The error message is

df.reindex(columns=['GDP'])
....
TypeError: values must be SparseArray

What I Hope to Achieve

Use reindex(columns=...) to filter selected columns contain encoded and non-encoded columns. Thanks!

Update (2018-01-17)

An issue is created at GitHub .

Answer 1

This certainly seems like a bug. As of v0.21, they've reworked a lot of their reindex API, so it seems something could've broken somewhere.

I don't have an answer, but I do have a workaround, hopefully it should do: You'll need to first transpose, and then reindex.

df.T.reindex(index=['GDP']).T

   GDP
0    1
1    2