How to create multiple series scatter plot with connected points using seaborn?

Question

I have a set of data stored in a pandas dataframe. I'm trying to use seaborn's pointplot() to create a multiple-series scatter plot with connected points. Each series has different (x,y) values, and they are stored as floats in my data frame. Each row has a label, differentiating each series. I'm using Python 2.7, seaborn version 0.5.1, and matplotlib version 1.4.3.

Everything I've managed to find tells me that I can achieve this with the following:

import matplotlib.pyplot as plt
import seaborn as sns

# Suppose my dataframe is called 'df', with columns 'x', 'y', and 'label'.
sns.pointplot(x = 'x', y = 'y', hue = 'label', data = df)

However, this results in some strange behavior:

• The colors are correctly identified, but only some of the points are connected
• The numbers on the x-axis overlap and it appears as though each data point is being labeled with it's value rather than scaling it with appropriate, clean values (seems to be treating the x data as a string/label rather than floats).

I attempted to work around this by splitting my data frame into pieces. This is not ideal because I may have about 10+ series to plot simultaneously, and I'd prefer to not split the data manually:

df1 = df[df.test_type.values == "label 1"]
df2 = df[df.test_type.values == "label 2"]

ax = sns.pointplot(x = 'x',y='y', color = "blue", data = df1)
sns.pointplot(x = 'x', y = 'y', data = df2, color="red", ax = ax)

In this case, all points are connected and they are colored appropriately, but again, the x axis is showing very strange behavior. Even though my x-values from each data frame are different, the plot aligns them so that they appear to be the same.

Now, I'm not sure how to post my output/plots cleanly, but some of my problems can be recreated with the following:

#import the necessary modules
import matplotlib.pyplot as plt
import pandas as pd
import seaborn as sns

#Here is some sample data. The 'x2' data is slightly offset from 'x1'
x1 = range(0,100,10)
x2 = range(1,100,10)
x = x1+x2

#The y-values I generate here mimic the general shape of my actual data
y1 = x1[::-1]
y2 = [i+25 for i in x1[::-1]]
y = y1+y2

#Two levels of labels that will be applied to the data
z1 = ["1"]*10
z2 = ["2"]*10
z = z1+z2

#A pandas data frame from the above data
df = pd.DataFrame({'x': x, 'y': y, 'z': z})

#Pointplot using the above data
sns.pointplot(x = 'x', y = 'y', data = df, hue = 'z')

Running this code results in the following:

• All x values across all series are spaced evenly. Note that the 'x2' values are the same as 'x1' translated by '1' and they are spaced at intervals of 10 within each series. I did not expect this behavior.
• The x axis doesn't have a "clean" looking scale. It literally labels each point's corresponding x-value. It labels the points correctly, but does not scale it appropriately. It seems like it's treating the x values as labels, similar to how a bar graph might behave.
• The points are colored correctly, but no points are connected.

To summarize my question:

Is there an easier/better/more elegant way to plot multiple-series scatter plots with connected points using data stored in a pandas data frame? Seaborn's pointplot looked ideal, but it is not functioning as I expected and I'm suspecting that it might serve a purpose different from what I need to accomplish. I'm open to other solutions that can achieve this (preferably using python).

Thanks in advance. I will update my question if I can figure out how to upload the output and plots from my code.

I'm 100% new to stackoverflow. I'd love to clarify my question by posting the plots being generated by my code, but I could not figure this out. Any pointers on how to do this would be much appreciated as well so I can update the question.

EDIT: It turns out that seaborn's pointplot uses the x-axis as a categorical axis, which explains the strange behavior I've mentioned above. Is there a way to manually change the x-axis behavior from categorical to numerical? This seems like the easiest approach but I'm not very familiar with fine-tuning plots in python.

Answer 1

I had a similar problem and I finally solved it using Seaborn's FacetGrid. I used plt.scatter for the points and the plt.plot for lines connecting the points.

g = sns.FacetGrid(df, hue="z", size=8)
g.map(plt.scatter, "x", "y")
g.map(plt.plot, "x", "y")

Note, this is done in Seaborn version 0.6.0 and version 0.5.1.

Answer 2

With the help of @mwaskom and this question, I've managed to find a solution to my posted question:

#Assuming df is a pandas data frame with columns 'x', 'y', and 'label'
for key,grp in df.groupby('label'):
    plt.plot(grp.x,grp.y,'o-',label = key)
plt.legend(loc = 'best')