python - Log-log lmplot with seaborn

40
votes

Can the function lmplot from Seaborn plot on a log-log scale? This is lmplot on a normal scale

import numpy as np
import pandas as pd
import seaborn as sns
x =  10**arange(1, 10)
y = 10** arange(1,10)*2
df1 = pd.DataFrame( data=y, index=x )
df2 = pd.DataFrame(data = {'x': x, 'y': y}) 
sns.lmplot('x', 'y', df2)

sns.lmplot('x', 'y', df2)

3
pythonseaborn
If you're not using facets, it will be easier to use seaborn.regplot. - mwaskom
I'm a little confused because your code draws two plots. Are you trying to reproduce the first plot with seaborn? Or do you want to draw the lmplot on top of the first plot? - mwaskom
@mwaskom the first plot should not be there. I remove it now from the code. - sjdh

3 Answers

65
votes

If you just want to plot a simple regression, it will be easier to use seaborn.regplot. This seems to work (although I'm not sure where the y axis minor grid goes)

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

x = 10 ** np.arange(1, 10)
y = x * 2
data = pd.DataFrame(data={'x': x, 'y': y})

f, ax = plt.subplots(figsize=(7, 7))
ax.set(xscale="log", yscale="log")
sns.regplot("x", "y", data, ax=ax, scatter_kws={"s": 100})

enter image description here

If you need to use lmplot for other purposes, this is what comes to mind, but I'm not sure what's happening with the x axis ticks. If someone has ideas and it's a bug in seaborn, I'm happy to fix it:

grid = sns.lmplot('x', 'y', data, size=7, truncate=True, scatter_kws={"s": 100})
grid.set(xscale="log", yscale="log")

enter image description here

5
votes

Call the seaborn function first. It returns a FacetGrid object which has an axes attribute (a 2-d numpy array of matplotlib Axes). Grab the Axes object and pass that to the call to df1.plot.

import numpy as np
import pandas as pd
import seaborn as sns

x =  10**np.arange(1, 10)
y = 10**np.arange(1,10)*2
df1 = pd.DataFrame(data=y, index=x)
df2 = pd.DataFrame(data = {'x': x, 'y': y})

fgrid = sns.lmplot('x', 'y', df2)    
ax = fgrid.axes[0][0]
df1.plot(ax=ax)        

ax.set_xscale('log')
ax.set_yscale('log')
4
votes

The simplest way to make a log-log plot from (probably) any seaborn plot is:

plt.xscale('log')
plt.yscale('log')

In the example:

import numpy as np
import pandas as pd
import seaborn as sns
x =  10**np.arange(1, 10)
y = 10** np.arange(1,10)*2
df1 = pd.DataFrame( data=y, index=x )
df2 = pd.DataFrame(data = {'x': x, 'y': y}) 
sns.lmplot('x', 'y', df2)
plt.xscale('log')
plt.yscale('log')

Link to resulting image