ipympl not working nicely with interact

Trying to follow some simple examples of Jupyter widgets with matplotlib, but I'm having issues when using the `ipympl` backend. The first update to a plot produces a second plot (overplotted), and the second update just erases the figure window altogether. This does not happen when using `inline`.

Here's some simple code to reproduce the issue:

%matplotlib ipympl
import matplotlib.pyplot as plt
from ipywidgets import interact

def f(n):
plt.plot([0,1,2],[0,1,n])
plt.show()
interact(f,n=(0,10));

If I use `inline`, this will update the line plot. With `ipympl` it fails.

I am using ipympl 0.1.0 from Anaconda, together will other packages:

ipywidgets 7.0.0 py36_intel_0 [intel] intel
ipympl 0.1.0 py36_1 conda-forge
matplotlib 2.0.2 np113py36_intel_1 [intel] intel
ipython 6.1.0 py36_intel_0 [intel] intel
jupyter 1.0.0 py36_intel_5 [intel] intel
jupyter_client 5.1.0 py36_intel_0 [intel] intel
jupyter_console 5.1.0 py36_intel_0 [intel] intel
jupyter_core 4.3.0 py36_intel_1 [intel] intel

Run the supplied ipympl and interact reproducer, then compare its figure-update behavior with the inline backend. Trace the ipympl backend and widget update path involved in repeated interact calls; done means updates keep one plot visible without erasing the figure.

Written by the indexing model from the issue text.