Section: Quantitative Data | Data Visualization on Seaborn

Collapse Expand
Definition
- Select activity Que sont les données numériques? Les données quant...
  
  What is quantitative data?
  
  Quantitative, or numerical, data, as the name suggests, consist of measurable values, that is, numbers.
  
  We will introduce and test the different types of charts that can be created with Seaborn.

Relplot

The relplot() function can be used to create scatter plots and line plots.

Here is the function signature:

Signature relplot

There is, of course, documentation available online, so we will only cover the most essential elements needed to display what we need as quickly as possible, namely:

Parameter name	Explanations	Required type	Example
data	Table you are planning to use	DataFrame, Series, dict, array, or list of arrays	data=table
x	Variable of the table for the x-axis	string corresponding to a variable of the table	x="weight"
y	Variable of the table for the y-axis	string corresponding to a variable of the table	y=”height”
hue	Variable of the table to distinguish with colors	string corresponding to a variable of the table	hue=”age”
size	Variable of the table to distinguish with different sizes	string corresponding to a variable of the table	size=”money”
style	Variable of the table to distinguish with different styles	string corresponding to a variable of the table	style=”sex”
row	Variable of the table that will be used to make different plots, in rows	string corresponding to a variable of the table	row=”category”
col	Variable of the table that will be used to make different plots, in columns	string corresponding to a variable of the table	col=”job”
kind	Type of plot desired	string, 2 choices	kind=”scatter” or kind=”line”

Here is an example using the following code, which you can run:

 Importation Code
import pandas as pd
import seaborn as sns
import numpy as np
import matplotlib.pyplot as plt
from pyodide.http import open_url

url = "https://raw.githubusercontent.com/mwaskom/seaborn-data/master/penguins.csv"
data = pd.read_csv(open_url(url))
← Running

 Cell 2
sns.relplot(data=data,x="bill_length_mm",y="bill_depth_mm",hue="species",style="sex",size="body_mass_g",col="island")← Running

It can be observed that col allows you to create different plots within the same figure, hue uses colors to distinguish one variable, and style uses the shape of the markers to distinguish another variable. You can experiment with different parameters and rerun the code.

Ellipses can be added to relplot scatter plots. To draw on the plot, you first need to retrieve the ax object:

 Cell 3
from matplotlib.patches import Ellipse

g = sns.relplot(
    data=data,
    x="bill_length_mm",
    y="body_mass_g",
    hue="species",
    kind="scatter",
    height=5
)
ax = g.ax

def add_confidence_ellipse(x, y, ax, n_std=2.0, **kwargs):
    # Delete NaN before calculation
    mask = ~(np.isnan(x) | np.isnan(y))
    x, y = x[mask], y[mask]
    if len(x) < 2:
        return
    cov = np.cov(x, y)
    mean = np.mean(x), np.mean(y)
    eigvals, eigvecs = np.linalg.eigh(cov)
    order = eigvals.argsort()[::-1]
    eigvals, eigvecs = eigvals[order], eigvecs[:, order]
    angle = np.degrees(np.arctan2(*eigvecs[:, 0][::-1]))
    width, height = 2 * n_std * np.sqrt(eigvals)
    ellipse = Ellipse(
        xy=mean,
        width=width,
        height=height,
        angle=angle,
        fill=False,
        **kwargs
    )
    ax.add_patch(ellipse)

palette = sns.color_palette()

for i, species in enumerate(data["species"].unique()):
    subset = data[data["species"] == species]
    add_confidence_ellipse(
        subset["bill_length_mm"].values,
        subset["body_mass_g"].values,
        ax,
        edgecolor=palette[i],
        linewidth=2
    )

ax.set_xlabel("Bill length (mm)")
ax.set_ylabel("Body mass (g)")
ax.set_title("95% Confidence Ellipse by Species — Penguins")← Running

Ellipse comes from matplotlib.patches.

You can also display line plots by setting kind to "line":

 Cell 4
sns.relplot(data=data,x="bill_length_mm",y="bill_depth_mm",hue="species",style="sex",col="island",kind="scatter")
← Running

size cannot be used with a plot of kind "line".
This time, style changes the line style according to the value of the "sex" variable.

Displot

The displot() function allows you to display different types of distributions.

signature displot

Parameter name	Explanations	Required type	Example
data	Table you are planning to use	DataFrame, Series, dict, array, or list of arrays	data=tableau
x	Variable of the table for the x-axis	string corresponding to a variable of the table	x="weight"
y	Variable of the table for the y-axis	string corresponding to a variable of the table	y=”height”
hue	Variable of the table to distinguish with colors	string corresponding to a variable of the table	hue=”age”
row	Variable of the table that will be used to make different plots, in rows	string corresponding to a variable of the table	row=”category”
col	Variable of the table that will be used to make different plots, in columns	string corresponding to a variable of the table	col=”job”
kind	Type of plot desired	string, 3 choices	kind=”hist”,kind=”kde” or kind=”ecdf”
rug	Allows to see individual values on the 2 axes	Boolean	rug=True

 Importation Code
import pandas as pd
import seaborn as sns
import numpy as np
import matplotlib.pyplot as plt
from pyodide.http import open_url

url = "https://raw.githubusercontent.com/mwaskom/seaborn-data/master/penguins.csv"
data = pd.read_csv(open_url(url))
← Running

Here is an editable example code that allows you to create a histogram:

 Cell 2
sns.displot(data=data, x="bill_length_mm", rug=True, hue="sex", bins=20)
← Running

If no data is specified for the y-axis, the y-axis will represent the number of occurrences. If kind is not specified, a histogram is displayed by default.

The bins argument controls the number of bars. The rug parameter allows you to visualize individual observations along the axes of the plot.

We can also use Kernel Density Estimation (KDE) to estimate a distribution. Here is an example of how to use it:

 Cell 3
sns.displot(data=data,x="bill_length_mm", rug=True, hue="sex", kind="kde")
← Running

If a variable is specified for y:

 Cell 4
sns.displot(data=data,x="bill_length_mm", y="bill_depth_mm", rug=True, hue="sex", kind="kde")
← Running

A plot of this type can be read like a contour map. Each line connects points with similar probability densities. The centers of the contours correspond to areas of higher density.

The last type of distribution available is the ECDF (Empirical Cumulative Distribution Function). The y parameter cannot be specified for this type of distribution, as it is univariate.

 Cell 5
sns.displot(data=data, x="body_mass_g", rug=True, hue="sex", kind="ecdf", row="species", col="sex", height=5)
← Running

The row parameter allows you to display even more plots based on another variable in the dataset. The data contains three penguin species, so there are three rows of plots. There are two sexes in the dataset, so there are two columns. height controls the height of the plots.

Boxplot and Violinplot

A common graphical representation of data is the box plot, which can be created using boxplot().

signature boxplot

Parameter name	Explanations	Required type	Example
data	Table you are planning to use	DataFrame, Series, dict, array, or list of arrays	data=tableau
x	Variable of the table for the x-axis	string corresponding to a variable of the table	x="weight"
y	Variable of the table for the y-axis	string corresponding to a variable of the table	y=”height”
hue	Variable of the table to distinguish with colors	string corresponding to a variable of the table	hue=”age”
dodge	Variable allowing to choose if the content overlap	Boolean	dodge=False
width	Variable allowing to control the width of the boxes	Float value	width=0.5
gap	Variable allowing to control the size of the gap between the "dodged" boxes	Float value	gap=0.1

Here is an example of code that you can modify and run:

 Importation Code
import pandas as pd
import seaborn as sns
import numpy as np
import matplotlib.pyplot as plt
from pyodide.http import open_url

url = "https://raw.githubusercontent.com/mwaskom/seaborn-data/master/penguins.csv"
data = pd.read_csv(open_url(url))
← Running

 Cell 2
sns.boxplot(data=data, x="bill_length_mm", hue="island", dodge=True, width=0.5)
← Running

By default, gap is set to 0. The orientation is handled automatically by Seaborn, but if the plot is two-dimensional with two numerical variables, it can be specified using orient, which can be set to "h" or "v".

 Cell 3
sns.boxplot(data=data, x="bill_length_mm", hue="island", dodge=True, width=0.5, gap=0.1, log_scale=True)
← Running

log_scale allows you to change the scale. A numerical value specifies the base, which is 10 by default. If the plot is two-dimensional, two values can be provided, one for each axis.

The violin plot is also available through violinplot().

signature violinplot

Parameter name	Explanations	Required type	Example
data	Table you are planning to use	DataFrame, Series, dict, array, or list of arrays	data=tableau
x	Variable of the table for the x-axis	string corresponding to a variable of the table	x="weight"
y	Variable of the table for the y-axis	string corresponding to a variable of the table	y=”height”
hue	Variable of the table to distinguish with colors	string corresponding to a variable of the table	hue=”age”
inner	Variable allowing to choose the inner representation of the violin	string corresponding to a type of representation	inner=”box”,inner=”quart”,inner=”point”
split	Variable allowing to choose if the violin will be symetric	Boolean	split=True
width	Variable allowing to control the width of the violin	Float value	width=0.5
dodge	Variable allowing to choose if the content overlap	Boolean	dodge=False
gap	Variable allowing to control the size of the gap between the "dodged" violins	Float value	gap=0.1

Here is an example of code to run:

 Cell 4
sns.violinplot(data=data, x="bill_length_mm", hue="sex", dodge=True, linewidth=3, split=True, inner="point")
← Running

split allows two distributions to be displayed on the same violin plot, as they are symmetrical. linewidth controls the thickness of the outline.

We displayed the individual data points inside the violin plot, but we can instead choose to display a miniature box plot using inner="box":

 Cellule 5
sns.violinplot(data=data, x="bill_length_mm", hue="sex",dodge=True,linewidth=3, split=True, inner="box")
← Running

Regplot and Lmplot

If you want to perform linear regressions, Seaborn provides a dedicated function: regplot().

Parameter name	Explanations	Required type	Example
data	Table you are planning to use	DataFrame, Series, dict, array, or list of arrays	data=table
x	Variable of the table for the x-axis	string corresponding to a variable of the table	x="weight"
y	Variable of the table for the y-axis	string corresponding to a variable of the table	y=”height”
ci	Variable allowing to control the confidence interval	Integer between 1 and 100	ci=99
nboot	Variable allowing to choose the number of bootstrap resampling done.	Integer	nboot=100
seed	Variable allowing reproductibility by choosing the seed of the bootstrap.	Integer	seed=42
logistic	Variable allowing to do a logistic regression	Boolean	logistic=True
lowess	Variable allowing to do a LOWESS regression	Boolean	lowess=True
robust	Variable allowing to do a robust regression. Higher computing cost	Boolean	robust=True

regplot() can also display the confidence interval around the regression line, which is set to 95% by default.

Here is an editable example of code:

 Importation code
import pandas as pd
import seaborn as sns
import numpy as np
import matplotlib.pyplot as plt
import statsmodels
from pyodide.http import open_url

url = "https://raw.githubusercontent.com/mwaskom/seaborn-data/master/penguins.csv"
data = pd.read_csv(open_url(url))
← Running

 Cell 2
sns.regplot(data=data, x="bill_length_mm",y="bill_depth_mm", ci=70)
← Running

Here, we are 70% confident that the true regression curve lies within the interval displayed on the graph. By default, n_boot is set to 1000. Increasing this value will necessarily increase the code's execution time, as additional resampling iterations will be performed. seed allows you to reproduce the same resampling results by using an integer as a seed, which is useful for reproducibility when writing a scientific paper or verifying that a method works correctly.

The regression method can also be changed by selecting a different approach, for example by setting the lowess parameter to True:

 Cell 3
sns.regplot(data=data, x="bill_length_mm", y="bill_depth_mm",
 ci=99, lowess=True)
← Running

The confidence interval is not displayed when using LOWESS.

Another option is lmplot(), which is better suited for performing regressions across multiple plots:

signature lmplot

Parameter name	Explanations	Required type	Example
data	Table you are planning to use	DataFrame, Series, dict, array, or list of arrays	data=tableau
x	Variable of the table for the x-axis	string corresponding to a variable of the table	x="weight"
y	Variable of the table for the y-axis	string corresponding to a variable of the table	y=”height”
hue	Variable of the table to distinguish with colors	string corresponding to a variable of the table	hue=”age”
row	Variable of the table that will be used to make different plots, in rows	string corresponding to a variable of the table	row=”category”
col	Variable of the table that will be used to make different plots, in columns	string corresponding to a variable of the table	col=”job”
ci	Variable allowing to control the confidence interval	Integer between 1 and 100	ci=99
nboot	Variable allowing to choose the number of bootstrap resampling done.	Integer	nboot=100
lowess	Variable allowing to do a LOWESS regression	Boolean	lowess=True

Here is an example of code:

 Cell 4
sns.lmplot(data=data, x="bill_length_mm", y="bill_depth_mm", ci=95, hue="island", robust=True, col="sex")
← Running

Robust and logistic regressions are also available, just as with regplot(). The nboot and seed parameters are also available.

Heatmap and Clustermap

Seaborn also allows you to create heatmaps using heatmap().

Parameter name	Explanations	Required type	Example
data	Table you are planning to use	DataFrame, Series, dict, array, or list of arrays	data=table
cmap	Heatmap colors: either a Matplotlib color palette or a custom palette.	String corresponding to a color palette or color_palette of Seaborn	cmap="viridis" or cmap=sns.color_palette("light:blue", as_cmap=True)
annot	Variable that determines whether the cell values are displayed.	Boolean	annot=True, False by default
vmin	Minimum value considered for the colormap.	Float value	vmin=30.6
vmax	Maximum value considered for the colormap.	Float value	vmax=42
linecolor	Variable used to choose the color of the lines between cells.	String corresponding to a color	linecolor="blue"
linewidths	Variable controlling the thickness of the lines between cells.	Float value	linewidths=0.2 or linewidths=10
mask	Variable used to control the values displayed in the heatmap.	Boolean table with the same format as data	mask=tableau_mask

Here is an example of code:

 Importation Code
import pandas as pd
import seaborn as sns
import numpy as np
import matplotlib.pyplot as plt
import scipy
from pyodide.http import open_url

url = "https://raw.githubusercontent.com/mwaskom/seaborn-data/master/glue.csv"
glue = pd.read_csv(open_url(url)).pivot(index="Model",columns="Task",values="Score")
← Running

 Cell 2
sns.heatmap(glue)
← Running

We use pivot to format the data in the desired order:

index specifies the variable for the y-axis.
columns specifies the variable for the x-axis.
values must be a numerical variable, and this is what the heatmap will use to determine the cell colors.

 Cell 3
sns.heatmap(glue,cmap="viridis",annot=True,vmin=20,vmax=80,linecolor="red",linewidths=0.1)
← Running

With the vmin and vmax parameters, we can define the range of values over which the heatmap will be applied. We also have graphical options such as linecolor and linewidths to customize the lines between cells. annot displays the values in each cell of the heatmap.

If you need to perform clustering on a heatmap, you can use Seaborn's clustermap(). Note that this function requires SciPy, so you will need to install it in the environment you are working in. If you are using Google Colab, this will not be necessary, as you can import it directly.

signature clustermap

Parameter name	Explanations	Required type	Example
data	Table you are planning to use	DataFrame, Series, dict, array, or list of arrays	data=table
method	SciPy method for clustering	String corresponding to a SciPy method	method="centroid"
metric	SciPy metric for clustering	String corresponding to a SciPy metric	metric="jaccard"
z_score	Variable used to standardize the data	0 for rows, 1 for columns	z_score=0
standard_scale	Variable used to normalize the data	0 for rows, 1 for columns	standard_scale=1
row_cluster, col_cluster	Variables used to choose the clustering axes	Boolean	row_cluster=False
figsize	Variable controlling the size of the figure	tuple(width,height)	figsize=(4,4)
dendrogram_ratio	Variable controlling the size ratio of the dendrograms	tuple(row ratio, column ratio)	dendrogram_ratio=(0.2,0.1)
cbar_pos	Variable controlling the position of the color bar	tuple(left,bottom,width,height)	cbar_pos=(0,0.1,0.05,0.6)

Here is an example of code:

 Cell 4
url = "https://raw.githubusercontent.com/mwaskom/seaborn-data/master/iris.csv"
iris = pd.read_csv(open_url(url))
species = iris.pop("species")

# Partie modifiable
sns.clustermap(iris)
← Running

We remove the extra variable using pop() to perform the clustering. In this case, the variable is "species", which we will reuse later.

The dendrograms are the trees displayed on the sides of the clustermap that represent the different clusters formed.

Now, let's explore some of the different parameters:

 Cell 5
lut = dict(zip(species.unique(), sns.color_palette("Set2", species.nunique())))
row_colors = species.map(lut)

sns.clustermap(
    iris,
    row_cluster=True,
    dendrogram_ratio=(0.3,0.1),
    row_colors=row_colors,
    method="weighted",
    metric="correlation",
    z_score=1,
    annot=True,
    figsize=(7,12),
    cbar_pos=(0,0.1,0.02,0.8)
)
← Running

row_cluster groups the rows according to their similarity in order to reveal clusters. dendrogram_ratio controls the size of the dendrograms; the first value corresponds to the one on the left and the second to the one at the top. row_colors adds a color next to the rows. Here, using the previous lines, the species corresponding to each row are displayed.

metric allows you to choose the similarity distance used, while method specifies the algorithm used to perform the clustering. Setting z_score to 1 indicates that the data is normalized across the rows. cbar_pos allows you to choose the position of the color bar. annot displays the values in each cell. figsize controls the size of the figure.

Quantitative Data

Section outline

Definition

What is quantitative data?

Relplot

Displot

Boxplot and Violinplot

Regplot and Lmplot

Heatmap and Clustermap