Seaborn Barplot with Specified Confidence Intervals

Question

I want to plot confidence intervals on a Seaborn barplot, but I already have the confidence intervals computed. How can I get Seaborn to plot my confidence intervals instead of trying to compute them itself?

For instance, suppose I have the following pandas DataFrame:

   x = pd.DataFrame([
        ['Group 1', 0.5, 0.05],
        ['Group 1', 0.6, 0.07],
    ], columns=['Group', 'Mean', 'SD'])

How do I plot a bar chart with those means and standard deviations?

Answer 1

You could use seaborn to draw a bar chart without error bars. And then use matplotlib's errorbar to add the error bars. The code below assumes the 'Group' column contains two different values:

from matplotlib import pyplot as plt
import seaborn as sns
import pandas as pd

x = pd.DataFrame([
    ['Group 1', 0.5, 0.05],
    ['Group 2', 0.6, 0.07],
], columns=['Group', 'Mean', 'SD'])

ax = sns.barplot(data=x, x='Group', y='Mean', color='dodgerblue')
ax.errorbar(data=x, x='Group', y='Mean', yerr='SD', ls='', lw=3, color='black')
plt.show()

Here is an attempt with nested bars. The errorbars are first drawn to get their standard positions and the x-coordinates are extracted. Then the bars get removed and created again with a new position. I am not sure whether it will work under all circumstances.

from matplotlib import pyplot as plt
import seaborn as sns
import pandas as pd
import numpy as np

x = pd.DataFrame([
    ['Group 1', 'A', 0.4, 0.08],
    ['Group 1', 'B', 0.5, 0.05],
    ['Group 1', 'C', 0.5, 0.02],
    ['Group 2', 'A', 0.6, 0.07],
    ['Group 2', 'B', 0.7, 0.09],
    ['Group 2', 'C', 0.7, 0.02],
    ['Group 3', 'A', 0.6, 0.07],
    ['Group 3', 'B', 0.2, 0.09],
    ['Group 3', 'C', 0.4, 0.11],
], columns=['Group', 'Subgroup', 'Mean', 'SD'])

num_hues = len(np.unique(x['Subgroup']))
ax = sns.barplot(data=x, x='Group', y='Mean', hue='Subgroup')
for (hue, df_hue), dogde_dist in zip(x.groupby('Subgroup'), np.linspace(-0.4, 0.4, 2 * num_hues + 1)[1::2]):
    bars = ax.errorbar(data=df_hue, x='Group', y='Mean', yerr='SD', ls='', lw=3, color='black')
    xys = bars.lines[0].get_xydata()
    bars.remove()
    ax.errorbar(data=df_hue, x=xys[:, 0] + dogde_dist, y='Mean', yerr='SD', ls='', lw=3, color='black')
plt.show()