Clean up jupyter notebooks

[terrorfisch]

• Automize stripping of metadata with a git filter. Possible tool nbtoolbelt.
• Use %matplotlib inline which includes a png instead of a javascript object
  • Pros:
  • smaller
  • no diffs due to version changes (the javascipt object depends on used versions)
  • Cons:
  • Fixed size

[qutech-lab]

Using %matplotlib inline has the minor drawback that the resulting plot cannot be interactively explored with the move and zoom UI offered by %matplotlib notebook. Another, potentially bigger, issue is that our plot function by default sets its show argument to True and thus issues a call to figure.show() which results in a warning when used in the %matplotlib inline context. This call to figure.show() is not necessary within a notebook as the figure will be displayed anyways. Suppressing is thus prevents the warning. It does, however, require us to either explicitely put show=False in each call to plot in our examples - which would be confusing since the plot is shown - or set show to False by default, which would require every use of plot outside of notebooks to explicitely pass show=True (or call the show() method on the returned figure object). Both of these do not really seem favorable to me.

-- lumip

[terrorfisch]

We can validate the output https://github.com/computationalmodelling/nbval

[terrorfisch]

Some other tools that might be useful:

• nbdime
• jq