How to annotate each lmplot facet by hue group or combined data

Question:

I’m trying to add annotations to lmplots in a FacetGrid (r and p values for each regression) but the plots have two regression lines because I’m using "hue", and therefore I get two annotations that are stacked on top of each other. I’d like to either specify that they are displayed in different locations or ideally to use the complete dataset, not separated by the argument passed to hue I assume for that I need to modify "data" in the annotate function but I cannot figure out how.

I did manage to do it by creating a dataframe that contains all r and p values and looping through g.axes_dict.items(), but I would like a more elegant solution where the values can be calculated and displayed directly

import pandas as pd
import seaborn as sns
import scipy as sp

dict = {
    'ID': ['A','B','C','D','A','B','C','D','A','B','C','D','A','B','C','D'],
    'SCORE': [18,20,37,40,34,21,24,12,34,54,23,43,23,31,65,78],
    'AGE': [34,54,46,65,43,23,54,23,43,54,23,32,56,42,12,43],
    'GENDER': [1,1,1,1,2,2,2,2,1,1,1,1,2,2,2,2]
}

df = pd.DataFrame(dict)

g = sns.lmplot(x='SCORE', y='AGE', data=df,hue='GENDER',
               col='ID', height=3, aspect=1)

def annotate(data, **kws):
    r, p = sp.stats.pearsonr(data['SCORE'], data['AGE'])
    ax = plt.gca()
    ax.text(.05, .8, 'r={:.2f}, p={:.2g}'.format(r, p),
            transform=ax.transAxes)

g.map_dataframe(annotate)

enter image description here

Asked By: user20501139

||
Source

Answers:

  • The tips dataset is being used because the sample data in the OP causes scipy to generate ConstantInputWarning: An input array is constant; the correlation coefficient is not defined.
  • Use a dict to define the y-position for each hue category
  • ideally to use the complete dataset
    • When using .map_dataframe, for each facet, each hue group is plotted separately, which can be seen by displaying data in def annotate.
    • If you are separating the data by using hue, then separate statistics should be plotted.
import seaborn as sns
import scipy


# function 
def annotate(data, **kws):

    # display data; see that for each Facet, hue groups are annotated separately - uncomment the following two lines
    # print(data.sex.unique())
    # display(data) 

    # get the hue group; there will be one
    g = data.sex.unique()[0]

    # get the y-position from the dict
    y = yg[g]

    r, p = scipy.stats.pearsonr(data['total_bill'], data['tip'])
    ax = plt.gca()
    ax.text(1, y, f'{g}: r={r:.2f}, p={p:.2f}')



# sample data
tips = sns.load_dataset('tips')

# define a y-position for each annotation in the hue group
yg = {'Male': 8, 'Female': 9}

# plot
g = sns.lmplot(x='total_bill', y='tip', col='time', data=tips, hue='sex', height=5, aspect=1)

# annotate
_ = g.map_dataframe(annotate)

enter image description here

Iterate through g.axes.flat

  • Alternative, do not use .map_dataframe.
  • Flatten and iterate through each axes, which easily allows for calculations and annotations to be made with all the data for each facet.
g = sns.lmplot(x='total_bill', y='tip', col='time', data=tips, hue='sex', height=5, aspect=1)

# flatten the axes for all the facets
axes = g.axes.flat

# iterate through each axes
for ax in axes:
    
    # get the title which can be used to filter the data by col
    col, group = ax.get_title().split(' = ')
    
    # select data from dataframe
    data = tips[tips[col].eq(group)]
    
    # get statistics
    r, p = scipy.stats.pearsonr(data['total_bill'], data['tip'])
    
    # annotate
    ax.text(2, 8, f'Combined: r={r:.2f}, p={p:.2f}')

Iterate through g.axes_dict.items()

  • This option has the col= groups as keys, but then hard coding 'time' is required for creating data.
g = sns.lmplot(x='total_bill', y='tip', col='time', data=tips, hue='sex', height=5, aspect=1)

# iterate through g.axes_dict
for group, ax in g.axes_dict.items():

    # select data from dataframe
    data = tips[tips['time'].eq(group)]
    
    # get statistics
    r, p = scipy.stats.pearsonr(data['total_bill'], data['tip'])
    
    # annotate
    ax.text(2, 8, f'Combined: r={r:.2f}, p={p:.2f}')

Plot Result

enter image description here

Answered By: Trenton McKinney