Support non-index dataframe in heatmap

[stevenlis]

Since Seaborn is planning to add support for dataframes other than pandas (#3369), I'd like to point out an issue with heatmap.

Currently, Seaborn's heatmap requires a Pandas dataframe with row labels or an index. However, certain dataframes like Polars do not have an index by design. It would be beneficial if the heatmap could provide an API that allows inputting a non-index dataframe.

import pandas as pd
import polars as pl
import seaborn as sns
import matplotlib.pyplot as plt

data = {
    'A': [1, 2, 3],
    'B': [2, 3, 4],
    'C': [1, 3, 5],
    'Index': ['I', 'II', 'III']
}

sns.heatmap(pd.DataFrame(data).set_index('Index'))

For polars, the current workaround:

• set ticklabels in matplotlib manually

df = pl.DataFrame(data)
fig, ax = plt.subplots()
sns.heatmap(df.drop('Index'), ax=ax)
ax.set_yticklabels(df.get_column('Index'))

• convert to pandas dataframe

sns.heatmap(pl.DataFrame(data).to_pandas().set_index('Index'))

[mwaskom]

I don't understand the request here. heatmap uses an index when it is available, but an index is not required, e.g. you can also pass a numpy array.

[stevenlis]

I'm not saying it doesn't work, but when you pivot a DataFrame in Polars, your index column will become just another column instead of an index object like in Pandas. This requires additional manipulation if you want to pass it to a seaborn heatmap, as shown above. I understand if this is not what Seaborn aims to cover.