seaborn: Selected KDE bandwidth is 0. Cannot estimate density

Question

import pandas as pd
import seaborn as sns

ser_test = pd.Series([1,0,1,4,6,0,6,5,1,3,2,5,1])
sns.kdeplot(ser_test, cumulative=True)

The above code g

Answer 1

if you don't want to wait for the seaborn git update to get released in a stable version, you can try one of the solutions in the issue page. specifically henrymartin1's suggestion to try manually passing in a small bandwidth inside a try/catch block (suggested by ahartikainen) which grabs the text of this specific error (so other errors still get raised):

try:
    sns.distplot(df)
except RuntimeError as re:
    if str(re).startswith("Selected KDE bandwidth is 0. Cannot estimate density."):
        sns.distplot(df, kde_kws={'bw': 0.1})
    else:
        raise re

This worked for me.

Answer 2

you have three options to try

first: showing KDE lumps with the default settings

sns.distplot(ser_test, hist = False, rug = True, rug_kws = {'color' : 'r'})

second: KDE with narrow bandwidth to show individual probability lumps

sns.distplot(ser_test, hist = False, rug = True, rug_kws = {'color' : 'r'}, kde_kws = {'bw' : 1})

third: choosing a different, triangular kernel function (lump shape)

sns.distplot(ser_test, hist = False, rug = True, rug_kws = {'color' : 'r'}, kde_kws = {'bw' : 1.5, 'kernel' : 'tri'})

Answer 3

pip uninstall statsmodels solved a similar problem with the same error.

Answer 4

What's going on here is that Seaborn (or rather, the library it relies on to calculate the KDE - scipy or statsmodels) isn't managing to figure out the "bandwidth", a scaling parameter used in the calculation. You can pass it manually. I played with a few values and found 1.5 gave a graph at the same scale as your previous:

sns.kdeplot(ser_test, cumulative=True, bw=1.5)

See also here. Worth installing statsmodels if you don't have it.

Answer 5

The problem occurs because of statsmodels.

Anyway, to solve the issue for seaborn version starting from 0.10.0, just place diag_kws={'bw': 1} as arg.

Try to figure out the optimal value for bandwidth.