pandas style can't set background gradient on multiindex after dataframe division

Question

I can do a pandas styles heatmap over a multi-index no problem:

df = sns.load_dataset('geyser').reset_index()

df['3m_duration'] = df.duration > 3

group_cols = ['kind', '3m_duration']

count_gpby = df[
    group_cols + ['index']
].groupby(
    group_cols
)

count_gpby.count().style.background_gradient(cmap ='Blues')

I can also divide a subset groupby by the total groupby to get a comparative rate/ratio per group:

df['binary'] = 'A'
df.loc[100:, 'binary'] = 'B'

subset_gpby = df[
    group_cols + ['index']
].loc[df.binary=='B'].groupby(
    group_cols
).count()

(subset_gpby / gpby).style.background_gradient(cmap ='Blues')

But then I try to combine these two "views" as two columns in the same multi-index dataframe so that I can see a raw count and a comparative ratio at the same time. This has no issue printing:

But it cant be displayed with the Pandas style heatmap background gradient because of "non-unique indices":

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-275-f82cbb6545e2> in <module>
----> 1 pd.concat([(subset_gpby / gpby), gpby], axis=1).style.background_gradient(cmap ='Blues')

C:\ProgramData\Anaconda3\envs\venv\lib\site-packages\pandas\core\frame.py in style(self)
    959         from pandas.io.formats.style import Styler
    960 
--> 961         return Styler(self)
    962 
    963     _shared_docs[

C:\ProgramData\Anaconda3\envs\venv\lib\site-packages\pandas\io\formats\style.py in __init__(self, data, precision, table_styles, uuid, caption, table_attributes, cell_ids, na_rep, uuid_len)
    161             data = data.to_frame()
    162         if not data.index.is_unique or not data.columns.is_unique:
--> 163             raise ValueError("style is not supported for non-unique indices.")
    164 
    165         self.data = data

ValueError: style is not supported for non-unique indices.

However,

pd.concat([(subset_gpby / gpby), gpby], axis=1).index.value_counts()

> (short, False)    1
> (short, True)     1
> (long, True)      1
> (long, False)     1
> dtype: int64

shows that there's only one instance of each index, and the index is equal to the previous indices that render with no problem:

pd.concat([(subset_gpby / gpby), gpby], axis=1).index == (subset_gpby / gpby).index

> array([ True,  True,  True,  True])

Why does this error appear?

Answer 1

In pandas, both the "Index", and "Columns" are of type pd.Index. For this reason, both axes can be referred to as an Index. The Styler object only works on uniquely indexed DataFrames (See other limitations here), this includes both dimensions.

When concating the two items we end up with multiple columns named 'index':

pd.concat([(subset_gpby / gpby), gpby], axis=1)

                      index  index  # <- Note the duplicate column names
kind  3m_duration                 
long  False        1.000000      1
      True         0.631579    171
short False        0.635417     96
      True         0.500000      4

Since we don't have meaningful column names, we can simply pass ignore_index=True to concat (Note this only affects the axis of concatenation, axis=1 in this case):

pd.concat([(subset_gpby / gpby), gpby], axis=1, ignore_index=True)

                          0    1
kind  3m_duration               
long  False        1.000000    1
      True         0.631579  171
short False        0.635417   96
      True         0.500000    4

Or, as always, we can rename columns to be meaningful. However, we need something like set_axis here since rename would affect all columns named "index":

pd.concat(
    [(subset_gpby / gpby), gpby], axis=1
).set_axis(['Subset gpby', 'gpby'], axis=1)

                   Subset gpby  gpby
kind  3m_duration                   
long  False           1.000000     1
      True            0.631579   171
short False           0.635417    96
      True            0.500000     4

Anyway we do it, we will be able to use background_gradient again since the column index is unique:

pd.concat(
    [(subset_gpby / gpby), gpby], axis=1, ignore_index=True
).style.background_gradient(cmap='Blues')

---

Setup used:

import pandas as pd

import seaborn as sns

# Setup Data
df = sns.load_dataset('geyser').reset_index()
group_cols = ['kind', '3m_duration']
df['3m_duration'] = df['duration'].gt(3)
subset_df = df[[*group_cols, 'index']].copy()
# Build Count DataFrames
gpby = subset_df.groupby(group_cols).count()
subset_gpby = subset_df.loc[100:, :].groupby(group_cols).count()