Next Steps

You started this course able to make a chart. You are leaving it able to choose the right chart — and to say what it reveals, what it hides, and when it would break. That shift, from calling functions to reasoning about data, is the whole point. Let's name what you now carry with you, and then look at where to go from here.

What you can now do

• Get the data shape right first. You reach for tidy/long data by default — one variable per column, one observation per row — and you know to melt a wide table before fighting with a plot. Most "why won't this work?" moments are a shape problem, not a parameter problem.
• Interrogate every chart. You ask the three questions of anything you draw: What does it reveal? What does it hide? When does it break? That habit catches misleading bin widths, overplotted scatters, and confounded groups before they reach a reader.
• Pick from the chart families on purpose. You know the map: relational plots (scatter, line) for how two numbers move together; distribution plots (histogram, KDE, ECDF, rug) for the shape of one; categorical plots (bar, count, box, violin, boxen, strip, swarm) for comparing groups; regression plots for trends; and multivariate grids (facets, pair plots, joint plots, heatmaps) for many variables at once.
• Choose figure-level vs axes-level deliberately. You reach for the figure-level functions (relplot, displot, catplot, lmplot, pairplot, jointplot) when you might want facets or a self-contained figure, and the axes-level ones (scatterplot, histplot, boxplot, heatmap, ...) when you are placing one chart on an Axes you control.
• Use aesthetics to communicate, not decorate. Themes, contexts, and color palettes are tools in service of a message — a diverging palette centered at zero, a message-style title, clear labels — not garnish you add at the end.

Where to go next

You have a solid foundation. Here are the most rewarding directions to take it.

The seaborn objects interface

Everything in this course used Seaborn's function interface — relplot, catplot, and friends. Seaborn also ships a newer, more composable objects interface, imported as seaborn.objects. Instead of choosing a function up front, you start a Plot describing your data and roles, then add layers onto it with a small, consistent grammar:

The idea is that so.Plot(...) sets up what you are plotting and .add(...) says how to draw it — and you can chain more .add(...) calls, statistical transforms, and scales to build a figure up piece by piece. It is the same statistical thinking you already have, expressed as a grammar of layers rather than a menu of functions. If you have met ggplot2 in R or a "grammar of graphics" before, this will feel familiar; if not, the mental model you built here transfers directly.

Deeper matplotlib, modeling, and your own data

• Matplotlib for total control. Seaborn draws on top of matplotlib, so when you need a custom annotation, an unusual layout, or pixel-level control, dropping down to the Axes and Figure objects underneath is always available. You touched this whenever you set a title or adjusted spacing.
• Statistical modeling for what plots only hint at. A chart can suggest a relationship; tools like statsmodels or scikit-learn let you quantify it — fit the regression, test whether a difference is real, estimate uncertainty. Visualization and modeling are partners: plot to form the hypothesis, model to test it, plot again to communicate it.
• Practice on your own data. The fastest way to cement all of this is to point it at a dataset you care about. Load it, run the EDA loop from the capstone — glance, question, chart, notice, repeat — and make one figure you would actually show someone.

Check your understanding

One last challenge

No new concept — just you, a dataset, and a chart of your choosing. Pick any built-in dataset (tips, mpg, or iris are all good), decide what question to ask, and draw the chart that answers it. The only requirement is that you assign your plot to g so the check can confirm you made something.

That is the loop you will run for the rest of your data life: get the shape right, ask what you want to see, draw it, read it honestly, and — when it matters — polish it for someone else. You have the foundations now. Go make pictures that tell the truth about your data.