Out of context and misleading instruction for saving figures

[hlapp]

The lesson states the following in section Saving your plots as image files:

If you'd like to save your plots as an image file, you can run fig.savefig('my_figure.png') where "my_figure" is the file name.

This won't work, and it's non-trivial to recover to a way that it does:

1. There is no variable fig here. Normally (in matplotlib documentation commonly found) this is the return value of plt.figure(), but we are not using this method here. Instead we use Seaborn.boxplot(), which does not return an equivalent value.
2. plt.savefig() would work (as one can find out from testing autocompletion alternatives), but appears to have to be in the same code cell as the one generating the plot. I.e., we can't just insert a new cell here and run this method.

[hlapp]

@dleehr FYI.

[choldgraf]

My 2 cents is that it'd be better to make sure that earlier on we create a fig variable, and then keep the fig.savefig method, as opposed to using plt.savefig. I think the former is more in-line with how Matplotlib "prefers".

[hlapp]

@choldgraf I agree with your argument, but am wondering how to apply this here where we are using the seaborn package.

There may well be an easy interface to obtaining the fig variable value, but I don't know enough about Python (or plotting in Python) to suggest how.

[choldgraf]

argg, I just looked through this notebook and it's actually using the non-object-oriented API for Matplotlib all the way through, which IMO is not recommended practice.

Would folks be OK with me taking a pass and changing the Matplotlib to use the object-oriented api instead of the pyplot api? E.g. doing:

fig, ax = plt.subplots()
ax.scatter(x, y)

instead of

plt.figure()
plt.scatter(x,y)

the former is recommended by the matplotlib community, while the latter is allowed more because a lot of Matlab people already want to plot with that API

[choldgraf]

(and to that extent, does anybody know what the _run notebook is for?

[hlapp]

I'd be all for moving the lesson to using recommended practices. I think the fraction of people in these workshops who are coming over from using Matlab is small. Also, other lessons in this curriculum, in particular the automation one, does I think use the object-oriented API, which was the source of some confusion among the students. (In fact I myself had been under the impression that the reason we're using the pyplot API is because it's what the Jupyter Notebook environment needs for plotting inline. Apparently that's false?)

[hlapp]

(and to that extent, does anybody know what the _run notebook is for?

It's the one now pointed to on Mybinder.org by the Sharing lesson, as the example for sharing an interactive live Jupyter Notebook. It's in essence a copy of the instructor's version of the notebook, with the interspersed and trailing exercises etc deleted.

Easy enough to recreate as a copy, I should add.