Seaborn: Statistical Data Visualization

Seaborn is a high-level Python data visualization library built on top of matplotlib. It is designed to create attractive and informative statistical graphics.

To begin using seaborn, you should follow these steps:

1. Install the package: Refer to the installation guide to download and set up the environment.
2. Learn the core concepts: Read the introductory notes to understand the library's design philosophy.
3. Explore examples: Browse the example gallery to see various plot types and implementations.
4. Consult the API: Use the API reference to find specific function signatures and parameters for your plotting tasks.

In version 0.3, regplot was changed to an "Axes-level" function, meaning it plots onto a specific set of matplotlib axes and no longer automatically produces marginal distributions.

To achieve the previous behavior (a bivariate plot with marginal distributions), use jointplot with the parameter kind="reg".

In version 0.7.0, stripplot defaults were updated to align more closely with swarmplot. Key changes include:

• Points are smaller by default.
• Points have no outlines by default.
• Points are not split by default when using the hue parameter.
• When using hue nesting with split=False, different hue levels are no longer drawn strictly on top of each other, improving visibility.

All these settings remain customizable via function parameters.

Seaborn v0.12.0 introduced the seaborn.objects interface, a new declarative, composable, and extensible API for statistical graphics inspired by the grammar of graphics (similar to ggplot2 and vega-lite).

Note: This interface is considered "experimental" and may undergo breaking changes in future minor releases. It is intended for users seeking a more structured way to compose plots by defining data mappings and transformations.

In heatmap and clustermap, automatic detection of diverging data has been removed. To use a diverging colormap, you must explicitly provide the center parameter.

When center is specified, the colormap is modified so that its middle color corresponds exactly to the center value. This ensures the colorbar limits correspond to the data range, though the full range of the colormap might not be used unless the data is symmetric around the center.

New perceptual uniform colormaps available in the seaborn.cm namespace include:

• Sequential: "rocket", "mako"
• Diverging: "icefire", "vlag"

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

data = np.random.randn(10, 12)
# Explicitly specify 'center' to trigger diverging colormap behavior
sns.heatmap(data, center=0, cmap="icefire")
plt.show()

When using kdeplot, the y-axis represents the probability density function (PDF), not the actual probability.

In a continuous probability distribution, the probability of any specific value is zero. The PDF is normalized so that the area under the curve (the integral) equals 1. If the data is concentrated in a very small range, the height of the curve (the density) must be greater than 1 to ensure the total area remains 1.

Seaborn provides several specialized interfaces and plot categories for statistical visualization. You can explore these through their respective tutorials and API references:

• Objects interface: The modern interface for composing graphics.
• Relational plots: For visualizing relationships between variables.
• Distribution plots: For visualizing the distribution of data.
• Categorical plots: For visualizing data categorized by discrete variables.
• Regression plots: For visualizing statistical relationships and regression models.
• Multi-plot grids: For creating complex layouts of multiple subplots.
• Figure theming: For controlling the overall aesthetics and styles of your figures.
• Color palettes: For managing color schemes and mapping data to colors.

The data processing engine has been refactored to support more flexible input formats:

• Pandas Indexing: Named variables in long-form data can now refer directly to a pandas.DataFrame index or multi-index levels without needing to call .reset_index() first.
• relplot Flexibility: Now accepts both long- and wide-format data, as well as data vectors in long-form mode.
• Dictionary Support: The data parameter can now accept a Python dict for wide-form data.
• Robustness: Wide-form data objects containing a mixture of types will now have non-numeric types removed automatically instead of raising an error.

Seaborn's categorical functions are designed for variables that take a finite number of non-numeric values (like strings).

Internal Mechanism: Historically, seaborn handles these by mapping unique values to 0-based integer indexes before passing them to matplotlib. This behavior occurs by default even if both $x$ and $y$ variables are numeric, which can cause unexpected behavior when mixing categorical and non-categorical plots.

Control with native_scale: Starting in version 0.13, you can use the native_scale parameter to control this:

• native_scale=False (default): Seaborn uses its internal integer mapping.
• native_scale=True: Seaborn preserves the original properties of the data used for categorical grouping, allowing for more predictable behavior with numeric data.