Direct plots and other objects using IPython protocol to notebook

[remram44]

TLDR %matplotlib inline for Vizier 😉

The "IPython protocol" is basically objects exposing _repr_html_(). IPython has a display() method that sends those to the notebook, and similarly does the same for the cell's "output" (last line).

To support everything that works in IPython:

• Use _repr_html_() to format rich objects for the notebook
• Need to make sure that calling IPython.display.display() sends objects to our frontend
  • ipykernel sets a DisplayPublisher (and InteractiveShell). By their own admittance, this part of IPython is not very well factored.
• Take the output from the last line of a Python cell and similarly run it through display()
• Check how matplotlib works (do they just call display()?)
  • ipykernel sets $MPLBACKEND to module://ipykernel.pylab.backend_inline with a show() that calls display(), so every figure created is sent to the notebook
  • Apache Zeppelin does the same thing
  • example

[okennedy]

This seems like it might be overkill. Admittedly, if matplotlib exposes this functionality and we can just plug it in to Vizier, that's great. However, doesn't VisTrails already have workflow steps for plotting?

[remram44]

Well we don't have to do all of this. Simply formatting "rich objects" by calling _repr_html_() should definitely be added since it is so easy.

Admittedly providing the IPython display() interface is a bit more involved, and might not be required, but I think that letting the users call matplotlib from their Python cells would be appreciated.

[okennedy]

+1

[okennedy]

@mrb24 Implemented something analogous to this already. Modulo documentation, perhaps this issue can be considered closed?