Drawing lines on scatter with Seaborn

Question

My objective is to draw a horizontal red line on y = 0 on a plot made using seaborn: sns.lmplotsplitted by col= or row=.

import numpy as np, seaborn as sns, pandas as pd
np.random.seed(5)

myData = pd.DataFrame({'x' :  np.arange(1, 101), 'y': np.random.normal(0, 4, 100),\
'z' : ['a','b']*50, 'w':np.random.poisson(0.15,100)})


sns.lmplot("x", "y", col="z", row="w", data=myData, fit_reg=False)
plt.plot(np.linspace(-20,120,1000), [0]*1000, 'r-')

We can see that only the last plot, of the array of plots, is marked by the red line:

Thanks for your help,

EDIT: reworded the question to account for the case where we generate an array of plots using col= and/or row= and we want the line to be traced on each plot.

Answer 1

So this chunk of code works for the general case where we use col=, row=, and hue=.

import numpy as np, seaborn as sns, pandas as pd
np.random.seed(5)

myData = pd.DataFrame({'x' :  np.arange(1, 101), 'y': np.random.normal(0, 4, 100),\
'z' : ['a','b']*50, 'w':np.random.poisson(0.15,100), 'hueMe':['q','w','e','r','t']*20})

myPlot = sns.FacetGrid(col="z", row='w', hue='hueMe', data=myData, size=5)
myPlot = myPlot.map(plt.scatter, "x", "y").set(xlim=(-20,120) , ylim=(-15,15))
myPlot = myPlot.map_dataframe(plt.plot, [-20,120], [0,0], 'r-').add_legend().set_axis_labels("x", "y")
plt.show()

Not sure why the color of the horizontal line comes out as the last color used on each individual plot, but I give up on this for now :)

Answer 2

Since I came across this looking for an answer, here is a more general answer that I eventually discovered:

map_dataframe will also accept a user defined function (and passes the data frame to this function) which is quite powerful because you can plot anything onto the facetgrid. In the OP case:

def plot_hline(y,**kwargs):
    data = kwargs.pop("data") #get the data frame from the kwargs
    plt.axhline(y=y, c='red',linestyle='dashed',zorder=-1) #zorder places the line underneath the other points

myPlot = sns.FacetGrid(col="z", row='w', hue='hueMe', data=myData, size=5)
myPlot.map(plt.scatter, "x", "y").set(xlim=(-20,120) , ylim=(-15,15))
myPlot.map_dataframe(plot_hline,y=0)
plt.show()

My problem was slightly more complex because I wanted a different horizontal line on each facet.

To replicate my case, assume the 'z' variable has two samples (a and b) and each with an observed value 'obs' (which I've added to myData below). 'hueMe' represents modeled values for each sample.

myData = pd.DataFrame({'x' :  np.arange(1, 101), 
                       'y': np.random.normal(0, 4, 100),
                       'z' : ['a','b']*50,
                       'w':np.random.poisson(0.15,100),
                       'hueMe':['q','w','e','r','t']*20,
                       'obs':[3,2]*50})

When you pass the data frame to plot_hline, you need to drop the duplicate values of 'obs' for each 'z' sample because axhline can only take a single value for y. (remember in our case each sample has 1 observed value 'obs' but multiple modeled 'hueMe' values). further, y must be a scalar (rather than a series) so you need to index into the data frame to extract the value itself.

def plot_hline(y,z, **kwargs):
    data = kwargs.pop("data") #the data passed in through kwargs is a subset of the original data - only the subset for the row and col being plotted. it's a for loop in disguise.
    data = data.drop_duplicates([z]) #drop the duplicate rows
    yval = data[y].iloc[0] #extract the value for your hline.
    plt.axhline(y=yval, c='red',linestyle='dashed',zorder=-1)


myPlot = sns.FacetGrid(col="z", row='w', hue='hueMe', data=myData, size=5)
myPlot.map(plt.scatter, "x", "y").set(xlim=(-20,120) , ylim=(-15,15))
myPlot.map_dataframe(plot_hline,y='obs',z='z')
plt.show()

resulting plot

Now seaborn maps the output from your function onto each facet of FacetGrid. Note, if you are using a different plotting function than axhline, you might not necessarily need to extract the value from the series.

Hope this helps someone!

Answer 3

Seaborn is really just an interface for matplotlib, so you can use all of your standard matplotlib functions as well. Importing pyplot and plotting a red horizontal line after your seaborn plot works for me.

import numpy as np, seaborn as sns, pandas as pd
import matplotlib.pyplt as plt
np.random.seed(5)

myData = pd.DataFrame({'x' :  np.arange(1, 101), 'y': np.random.normal(0, 4, 100)})

sns.lmplot("x", "y", data=myData, line_kws={'xdata': '0,1','ydata': '0,0','color': 'k', 'linestyle':'-', 'linewidth':'5'}, fit_reg=False)
plt.plot(np.linspace(-20,120,1000), [0]*1000, 'r')

My image is here - http://i.imgur.com/J7Lvt52.png