inconsistent plot between matplotlib and seaborn i

问题:

i am trying to add errorbars using plt.errorbar to a pointplot in seaborn:

import matplotlib
import matplotlib.pylab as plt
import seaborn as sns
import pandas
sns.set_style("white")

data = pandas.DataFrame({"x": [0.158, 0.209, 0.31, 0.4, 0.519],
                        "y": [0.13, 0.109, 0.129, 0.250, 1.10],
                        "s": [0.01]*5})

plt.figure()
sns.pointplot(x="x", y="y", data=data)
plt.errorbar(data["x"], data["y"], yerr=data["s"])
plt.show()

however the two plots look totally different even though the identical data is being plotted. what explains this and how can errorbars be added to a pointplot?

回答1:

It seems that sns.pointplot simply uses an array of [0...n-1] as x values and then uses the x values that you provide to label the ticks on the x-axis. You can check that by looking at ax.get_xlim() which outputs [-0.5, 4.5]. Therefore, when you provide the actual x values to plt.plot they just seem to be at a wrong position.

I wouldn't say that this is a bug, since seaborn considers the input to pointplot to be categorial (here's the documentation for more info)

You can solve this for your simple example by mimicking seaborns behavoir:

import matplotlib
import matplotlib.pylab as plt
import seaborn as sns
import pandas
import numpy as np
sns.set_style("white")

data = pandas.DataFrame({"x": [0.158, 0.209, 0.31, 0.4, 0.519],
                        "y": [0.13, 0.109, 0.129, 0.250, 1.10],
                        "s": [0.05]*5})

plt.figure()
sns.pointplot(x="x", y="y", data=data)
plt.errorbar(np.arange(5), data["y"], yerr=data["s"], color='r')
plt.show()