Skip to content


Subversion checkout URL

You can clone with
Download ZIP

Scipy 2013 BOF Notes

Michael Droettboom edited this page · 2 revisions

The original notes were taken on etherpad. @mdboom has reorganized them and summarized them here.

Interactive tweaking of plots


  • Interactive modification of plots would address the issues on tweaking commented on at the plotting competition

  • Need a way to write out the tweaks "as code" afterward to paste back into a script (or into the IPython notebook)

  • Igor Pro has a good track record of having 1-1 correspondence between image and code

  • Veusz can also produce code from a manually tweaked plot


  • vispy: High performance plotting project (in early days)

  • We should create a MEP on openGL (and more generally for high performance plotting)

  • An analysis of performance bottlenecks should be done, perhaps a standard set of benchmarks to track over time (#2188)

    • in general, this kind of a benchmarking of the fall-off of performance would be useful to point users to.

    • Experiment with reducing quality during panning/zooming, and then going back to full quality on idle (much like older 3D packages used to show only a bounding box during rotation). (#2189)

  • We should add built-in instrumentation of matplotlib methods -- this would allow us to collect statistics about performance as well as more easily help users debug problems when they are hard to reproduce. (#2190)

Continuous Integration


  • Use something we either rent or control instead of travis in order to collect more data on different platforms.

  • We need a larger test matrix -- Python, Numpy, GUI framework, Freetype and libpng versions. From this we can keep on top of our version requirements.

  • In a follow-up breakout session, we thought the minimum requirements would be:

    • Mac and Linux virtual machines that a number of core developers can log into

    • Windows virtual machines if feasible

    • Github integration (like what Travis does)

    • The ability for other machines with different configurations to publish their results to a public web server

    • Saving the output of the tests

    • Saving build products ("daily", or more often)

    • Auto-building the docs

  • We have Launchpad daily builds



  • use notebook output and potentially store the output in the notebook to store the image output in the repo

  • have an interactive plotting documentation system in which you can edit plots online (Google App engine)

Constraint-based layouts


  • Enthought uses the casuarius constraint solver for their (Enaml)[] layout engine -- steal or borrow?

  • Tight-layout goes a long way already - thank you jae-joon

  • See #1109 for earlier (and continued) discussion

  • There is disagreement about whether constrained layout and plot editor are orthogonal and whether we could pursue both in parallel or should pick one or the other as a focus

Matplotlib census

We should revive this.

survey results

"Plottable" objects


  • Plot method could take "plottable" objects (rather than just Numpy arrays), and those objects could obtain or generate data in other ways (such as for functional plotting)

  • This lead to a discussion about 2.0 (i.e. breaking some backward compatibility). However, it was noted that we were a room full of developers, not users.


  • We should keep track of the versions of Numpy and other dependencies in the major long-term-release Linux distributions (Debian, RedHat, Ubuntu) and ensure (probably though better CI) that we continue to support them. (#2191) The census (above) may help us with this kind of thing as well.
Something went wrong with that request. Please try again.