matplotlib: can I create AxesSubplot objects, then add them to a Figure instance?

For line plots, you can deal with the Line2D objects themselves:

fig1 = pylab.figure()
ax1 = fig1.add_subplot(111)
lines = ax1.plot(scipy.randn(10))

fig2 = pylab.figure()
ax2 = fig2.add_subplot(111)
ax2.add_line(lines[0])

The following shows how to "move" an axes from one figure to another. This is the intended functionality of @JoeKington's last example, which in newer matplotlib versions is not working anymore, because axes cannot live in several figures at once.

You would first need to remove the axes from the first figure, then append it to the next figure and give it some position to live in.

import matplotlib.pyplot as plt

fig1, ax = plt.subplots()
ax.plot(range(10))
ax.remove()

fig2 = plt.figure()
ax.figure=fig2
fig2.axes.append(ax)
fig2.add_axes(ax)

dummy = fig2.add_subplot(111)
ax.set_position(dummy.get_position())
dummy.remove()
plt.close(fig1)

plt.show()

Typically, you just pass the axes instance to a function.

For example:

import matplotlib.pyplot as plt
import numpy as np

def main():
    x = np.linspace(0, 6 * np.pi, 100)

    fig1, (ax1, ax2) = plt.subplots(nrows=2)
    plot(x, np.sin(x), ax1)
    plot(x, np.random.random(100), ax2)

    fig2 = plt.figure()
    plot(x, np.cos(x))

    plt.show()

def plot(x, y, ax=None):
    if ax is None:
        ax = plt.gca()
    line, = ax.plot(x, y, 'go')
    ax.set_ylabel('Yabba dabba do!')
    return line

if __name__ == '__main__':
    main()

To respond to your question, you could always do something like this:

def subplot(data, fig=None, index=111):
    if fig is None:
        fig = plt.figure()
    ax = fig.add_subplot(index)
    ax.plot(data)

Also, you can simply add an axes instance to another figure:

import matplotlib.pyplot as plt

fig1, ax = plt.subplots()
ax.plot(range(10))

fig2 = plt.figure()
fig2.axes.append(ax)

plt.show()

Resizing it to match other subplot "shapes" is also possible, but it's going to quickly become more trouble than it's worth. The approach of just passing around a figure or axes instance (or list of instances) is much simpler for complex cases, in my experience...