TODO: migrate into a blog article for reusability

The pyplot interface for Matplotlib is more of a helping tool for users who are already adept in plotting with MATLAB and would like to transition to Python-based tooling with minimal friction. The plotting commands are declarative: you call plot(x_arr, y_arr) to generate the plot, xlim(x_min, y_min) to set the x-axis limits, and savefig("filename.png") to generate the file. Like MATLAB, a GUI pops up when you call the plot function where you can manipulate and save the figure. The approach puts a strong emphasis to exploratory analysis and quick prototyping, at the cost of having the plotting defaults making your figures look dull.

Thought list:

• The goal is not to make the figures quickest to make or look the fanciest. We intend to strike a balance between writing terse code and pleasant data visualization.
• Matplotlib also promotes using its OOP-style approach to plotting.
• Common approach: data prototyping in the notebook, no need to save the figure; just keep reloading and re-running the cells.
• Always start with the subplots to generate both the figure and axis objects: it already makes it easy to add multiple axes within the figure, and allows you to resize it right within the generator.
• For quick prototyping, ax.set_<property> is sufficient, but a cleaner approach would be to use ax.set(**style_dict).
• If the figure components are not flexible enough to configure within the function call (e.g. violin plot), you can access each component within the axes and change them individually.
• It’s always faster to just use the pyplot interface, but a little effort to get used with object-based interface goes a long way, especially moving from exploration to presentation.

Examples: (Note: might be a separate post, mostly boilerplates or design patterns)

• NxN 2D projection of correlates in N-dimensions (joint prob.)
• Combine scatter plot with a 2D density plot
• Time-series violin plots
• Fitting a curve with custom confidence bands