Matplotlib 2.1.2直接创建Figure失败，plt.subplots()成功的原因咨询

Great question—this behavior makes sense once you understand how Matplotlib's module structure and initialization worked in older versions like 2.1.2. Let's break it down:

Root Cause: Top-Level matplotlib Doesn't Auto-Import All Submodules #

By default, when you run import matplotlib as mpl, the top-level matplotlib module doesn't automatically import its submodules (like matplotlib.figure) into its namespace. This was a deliberate design choice to keep startup times fast and avoid loading unnecessary code until it's needed.

So when you try to access mpl.figure.Figure() right after importing matplotlib, Python throws an AttributeError because figure isn't an attribute of the matplotlib module yet.

Why Importing a Backend Fixes This #

Nearly all Matplotlib backends (like backend_tkagg) depend on core components of the library—including the Figure class. When you import a backend, its code will explicitly import matplotlib.figure (or specific classes from it) to function properly.

In Python, when a submodule (like matplotlib.figure) is imported anywhere in your code, it gets added as an attribute to its parent module (matplotlib). That's why after importing a backend, mpl.figure suddenly becomes accessible.

Why Pyplot Works Without Explicit Backend Imports #

The pyplot interface is designed to handle all the tedious initialization steps for you. When you import matplotlib.pyplot, it automatically:

• Sets up a default backend (even if you don't specify one)
• Imports core submodules like matplotlib.figure, matplotlib.axes, etc.
• Configures the plotting environment

This is why plt.subplots() works immediately—pyplot has already done the work of importing the necessary modules and setting up the backend.

Is This Expected Behavior? #

Yes, this was intentional behavior in Matplotlib 2.1.2. The library was structured to let users pick and choose which parts to load, rather than forcing everything into memory upfront.

If you want to create a Figure without using pyplot or importing a backend, the most straightforward way is to import the Figure class directly:

from matplotlib.figure import Figure
fig = Figure()

This will work even without any backend imports, since you're accessing the class directly from its submodule.

内容的提问来源于stack exchange，提问作者tomanizer