Seaborn Object

Since the release of Seaborn 0.12, the Seaborn Objects API has been introduced. It provides a powerful alternative to the original plotting functions. This API is inspired by ggplot2 in R.

We will use a simple example. First, we import the objects as follows:

import seaborn.objects as so

so.Plot()

To this function, we specify the data that we are going to use:

so.Plot(tips,x=”total_bill”)

Here, tips is Seaborn's tips dataset.

Once the data has been specified, we can decide what to do with it using add(). Here, we create a histogram:

so.Plot(tips,x=”total_bill”).add(so.Bar(),so.Hist()).show()

And here is the result:

The equivalent of the scatterplots that can be created with relplot() is the Dot() object. After the import cell, all the code will be editable.

color plays the same role as the hue parameter introduced earlier, allowing the data to be differentiated according to a variable. marker allows another variable to be used to differentiate observations using different marker types, similarly to the style parameter. facet() serves a similar purpose to row and col, allowing the data to be distributed across multiple subplots. Finally, limit allows the display to be restricted to specific ranges on the x and/or y axes.

We can also easily add a regression curve using Line() and PolyFit():

This same Line() can have different types, such as PolyFit(), but it can also be used as a representation of the data:

If no type is specified for Line(), the data points are connected with lines. An interesting feature of using pandas DataFrames, as returned by load_dataset(), is that we can use .query() to perform SQL-like queries to select specific data. Here, we select only diamonds whose cut is Ideal and whose color belongs to a specific set of colors. The chained pipe() function passes this selected DataFrame as an argument to Plot(), after which other arguments such as x, y, and linestyle can be specified. The plotted line does not correspond to each individual observation. Indeed, using Agg() allows the data to be aggregated: each price value for a given depth is aggregated and averaged in the plot. The Band() and Est() objects can be used to display uncertainty around the curves.

The Path() object is an alternative to Line(), particularly suited for representing trajectories, as it connects the data points in the order in which they are presented.

If we want to display the area under curves, we can use Area(). The wrap parameter of facet() allows us to choose how many plots are displayed per row.

The Range() object allows intervals to be displayed and requires either bounds or an Est() object to calculate what should be displayed. With the latter, the mean and its confidence interval are displayed. Bounds can also be explicitly provided to define the interval to be displayed.

To create histograms, we use Bar() together with Hist(). We can choose the type of statistic to use. By default, "count" is used, but we can choose "density" for probability densities, "percent" for percentages, "probability" for proportions, or "frequency" for frequency.

Bar() can also be used to display, for example, a mean with Agg(), which allows data to be aggregated. Dodge() performs the same function as the dodge parameter in non-object-based plots.

To simply count occurrences, we can also use Bar(), combined with Count().

Seaborn objects can also be used to display percentiles with Perc(). We can choose which percentiles we want to display, and here we display them as Dot(). If no specific percentiles are selected, the default percentiles are [20, 40, 60, 80, 100].

We can also normalize the values using Norm(). Here, we normalize the values relative to the minimum year, which is 1970.

The Dot(), Line(), Path(), and Bar() objects have variants (Dots, Lines, etc.) that are better suited to large datasets.