How to normalize histograms in joinplot margins

Question

How can I normalize the histograms in sns.jointplot?

Right now I’m getting this plot

However, instead of showing 0, 200, 400 on the axis, I would like it to be a fraction of the total number of the dataset

This is my code:

import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
from matplotlib import rcParams
sns.set(style='white')

# sample x & y: doesn't match the top sample image
x = np.random.exponential(70, 100)
y = np.random.exponential(0.005, 100)

g = sns.jointplot(x, y, xlim = [30, 800], ylim = [0.0007, 0.023], marker = '*', s = 10, color = 'k',
             ec = 'k', marginal_kws=dict(bins=10, fill = True, log_scale=True), 
              height=1, ratio=6, marginal_ticks=True)
g.plot_joint(sns.kdeplot, color="r", zorder=0, levels=10, fill = True, thresh = 0.1)#, cmap = 'jet')

g.ax_joint.invert_xaxis()
g.fig.set_size_inches(14,10)
g.ax_joint.axvline(x = 70, color = 'red', ls = '--', lw = 2)#, label = '70 K')
g.ax_joint.axvline(x = 650, color = 'red', ls = '--', lw = 2)#, label = '70 K')
g.ax_joint.axhline(y = 0.005, color = 'red', ls = '--', lw = 2)#, label = '70 K')

Asked By: JohnGoodWill

||

Source

Answer 1

You can use g.ax_marg_x and g.ax_marg_y to access the axes objects corresponding to the marginal distributions. Then you can iterate over Axes.get_children() to filter out the "bar" objects which are represented as matplotlib.patches.Rectangle. Then the height/width of these rectangle objects can be adjusted and the axes limits can be adjusted as well:

import matplotlib.patches


def normalize_marginal(joint_grid, dim):
    get_size = lambda bar: getattr(bar, f'get_{"height" if dim == "x" else "width"}')()
    set_size = lambda bar, val: getattr(bar, f'set_{"height" if dim == "x" else "width"}')(val)
    set_lim = lambda ax, val: getattr(ax, f'set_{"y" if dim == "x" else "x"}lim')(val)

    ax = getattr(g, f'ax_marg_{dim}')

    bars = [obj for obj in ax.get_children() if isinstance(obj, matplotlib.patches.Rectangle)]
    sizes = [get_size(bar) for bar in bars]

    for bar, size in zip(bars, sizes):
        set_size(bar, size / sum(sizes))

    set_lim(ax, [0, 1])  # customize (e.g. `set_lim(ax, [0, 1.2*max(sizes)])`)


normalize_marginal(g, 'x')
normalize_marginal(g, 'y')

Answered By: a_guest

Answer 2

As stated in seaborn.jointplot, marginal_kws should accept parameters from the type of plot in the margins (e.g. sns.histplot in this case).
From sns.histplot: stat='probability': or proportion: normalize such that bar heights sum to 1.
- Pass this parameter to marginal_kws
If you want the total area of the histogram to equal 1, then use 'density'.
If there are multiple groups of data from using hue, then also consider adding common_bins=False and common_norm=False to marginal_kws.
Tested in python 3.10 and seaborn 0.11.2

import seaborn as sns
import numpy as np

# test data
np.random.seed(365)
x = np.random.exponential(70, 100)
y = np.random.exponential(0.005, 100)

# plot
g = sns.jointplot(x=x, y=y,  marker='*', s=10, color='k', height=7, ec='k', marginal_ticks=True,
                  marginal_kws=dict(bins=10, fill=True, log_scale=True, color='r', stat='probability'))
g.plot_joint(sns.kdeplot, color="r", zorder=0, levels=10, fill = True, thresh = 0.1)
g.ax_joint.invert_xaxis()

Instead of using xlim = [30, 800] and ylim = [0.0007, 0.023], which truncates the plot, you can mask the data first.
- Both masks are applied to x and y to ensure the resulting datasets are the same length.
This is optional, depending on how the data should be presented, and the bins will be normalized only on the selected data.

# test data
np.random.seed(365)
x = np.random.exponential(70, 100)
y = np.random.exponential(0.005, 100)

# masks
x_mask = (x > 30) & (x < 800)
y_mask = (y > 0.0007) & (y < 0.023)

# selected data
x_sel = x[x_mask & y_mask]
y_sel = y[x_mask & y_mask]

# plot
g = sns.jointplot(x=x_sel, y=y_sel,  marker='*', s=10, color='k', height=7, ec='k', marginal_ticks=True,
                  marginal_kws=dict(bins=10, fill=True, log_scale=True, color='r', stat='probability', common_bins=False, common_norm=False))
g.plot_joint(sns.kdeplot, color="r", zorder=0, levels=10, fill = True, thresh = 0.1)
g.ax_joint.invert_xaxis()

Answered By: Trenton McKinney

How to normalize histograms in joinplot margins

Question:

Answers: