5 Best Ways to Make a Stripplot with Jitter in Altair Python

by

💡 Problem Formulation: When visualizing data distributions across different categories, it’s common to use a stripplot. However, overlapping points can obscure patterns. To overcome this, jitter can be applied, which adds random noise to position of each point, separating them for better visibility. This article focuses on creating a stripplot with jitter in Altair, a declarative statistical visualization library for Python. We aim to show how to represent data such as a set of different species measurements in a way that each point, representing an individual measurement, is clearly distinguishable.

Method 1: Using the Mark Circle with Jitter Transform

Creating a stripplot with jitter in Altair can be achieved by combining mark_circle() with a transform that applies a random jitter. Here, the transform_jitter() function is specified, which allows us to control the extent of jittering along the x-axis for a clearer view of clustered points.

Here’s an example:

import altair as alt
from vega_datasets import data

source = data.iris()

stripplot = alt.Chart(source).mark_circle(size=8).encode(
    x=alt.X('species:N', axis=alt.Axis(title=None)),
    y='sepalLength:Q'
).transform_jitter(
    jitter_width=0.3
).properties(
    width=180,
    height=180
)

stripplot.show()

The output is a stripplot that displays sepal length measurements across different iris species, with points jittered along the x-axis.

In this snippet, the transform_jitter() is applied with a width of 0.3, providing a good amount of separation among the points without making the plot noisy. We’ve chosen to encode the species on the x-axis and sepal length on the y-axis. As a result, the data points are spread out horizontally within each species category, making overlapping points from the dataset more distinguishable.

Method 2: Adjust Jitter with a Random Number Generator Seed

The randomness of jitter can be controlled using a seed for the random number generator in the transform_jitter() function. By setting a seed, we can ensure that the jittered points are reproducible, which is key for creating consistent visualizations.

Here’s an example:

import altair as alt
from vega_datasets import data

source = data.iris()

stripplot = alt.Chart(source).mark_circle(size=8).encode(
    x='species:N',
    y='sepalLength:Q'
).transform_jitter(
    jitter_width=0.3, seed=1
).properties(
    width=180,
    height=180
)

stripplot.show()

The output is similar to Method 1 but with reproducible jitter due to the set seed.

We’ve specified a seed value in the transform_jitter() function as an assurance of consistency. If you want to present the same visualization again or share your work with others, setting a seed guarantees that the point spread stays the same every time the code is executed.

Method 3: Faceted Stripplot with Jitter to Compare Subgroups

Faceting is a valuable way to compare subgroups in your data. With Altair, you can create a faceted stripplot with jitter using columns or rows in the facet() channel. This method divides your stripplot into a matrix of small multiples, allowing for easy comparison across another categorical dimension.

Here’s an example:

import altair as alt
from vega_datasets import data

source = data.iris()

stripplot = alt.Chart(source).mark_circle(size=8).encode(
    x=alt.X('species:N', axis=alt.Axis(title=None)),
    y='sepalLength:Q',
    column='species:N'
).transform_jitter(
    jitter_width=0.3
).properties(
    width=180,
    height=180
)
stripplot.show()

The output is a series of stripplots, each representing a different iris species, enabling clear comparisons between groups.

By including the column='species:N' encoding, we’re asking Altair to create separate stripplots for each species laid out in a horizontal line. This method pairs the utility of jittering with the comparative power of faceting, enhancing analysis for distinct species groups.

Method 4: Jitter with Different Marks

Altair not only allows you to jitter points but also other shapes. This can be particularly important for emphasizing differences in data. Using mark_point() or mark_square() instead of mark_circle(), you can differentiate your jittered data points based on shape as well as position.

Here’s an example:

import altair as alt
from vega_datasets import data

source = data.iris()

stripplot = alt.Chart(source).mark_square(size=8).encode(
    x='species:N',
    y='sepalLength:Q'
).transform_jitter(
    jitter_width=0.5
).properties(
    width=200,
    height=180
)

stripplot.show()

The output shows a stripplot with square marks instead of circles, jittered across different species.

The use of mark_square() instead of mark_circle() provides a unique visual style and could serve to make specific data points stand out, depending on the context. Altering the shape of marks can help in tailoring the plot aesthetics or when making distinctions between multiple datasets on the same plot.

Bonus One-liner Method 5: Implementing Jitter Inline

For the quickest solution, Altair’s innate flexibility lets you add jitter directly within the encode() method of your chart definition. This one-liner is less customizable than using transform_jitter(), but it’s perfect for a fast, effective jitter.

Here’s an example:

import altair as alt
from vega_datasets import data

source = data.iris()

stripplot = alt.Chart(source).mark_circle(size=8).encode(
    x='jitter(species, width=0.3):O',
    y='sepalLength:Q'
).properties(
    width=180,
    height=180
)

stripplot.show()

The output is a compact stripplot with jitter applied directly through the encoding.

This technique saves time by integrating the jitter directly into the x-axis encoding of the chart, still providing a clear view of the distribution for each category, but with less configuration overhead. It’s an elegant shortcut that keeps your code concise.

Summary/Discussion

  • Method 1: Using Mark Circle with Jitter Transform. Strengths: Provides control over the jittering effect. Weaknesses: Requires an additional transform step.
  • Method 2: Adjust Jitter with a Random Number Generator Seed. Strengths: Guarantees reproducibility for consistent visualizations. Weaknesses: Less randomness can sometimes fail to distinguish very dense data points.
  • Method 3: Faceted Stripplot with Jitter to Compare Subgroups. Strengths: Allows comparison across subgroups through faceting. Weaknesses: Can be complex with numerous categories.
  • Method 4: Jitter with Different Marks. Strengths: Offers additional visual differentiation through mark shapes. Weaknesses: May be subjectively less clear for some audiences.
  • Bonus Method 5: Implementing Jitter Inline. Strengths: Fast and straightforward with minimal code. Weaknesses: Offers less control than the dedicated jitter transform.