Don't miss the links to Binder demo and further reading at the final slide.
The video of the talk is out! Watch it:
Alt: Youtube link
Rewind of my experience
I got to meet a lot of great like-minded people in those two days! It is no easy task to list all whom I met and convey my regards. Therefore, I list the videos and slides which might be of interest to you as a reader:
- Erik Sundell who works for both Jupyterhub and his company shared his insights on how one can setup Jupyter with multiple user deployments on a single server or on the cloud. [video]
- Gavin Chan who compared Cython and Pybind11 in his talk. [video]
- Isaac Bernat showed us that algorithmic improvements go a long way, indeed! [video]
There were also really educative talks on asyncio, property based testing [slides] and mutation testing, plus a sizable representation from the data science community. A couple of fun projects such as the a plotting DIY tool made using Python + Raspberry Pi Zero, BrachioGraph and a really hacky self-documenting code. See the PyCon Sweden video channel for the whole programme.
Last but not the least, I am thankful to the whole PyCon Sweden team for selflessly devoting their time into organizing a great conference - twice as large compared to last year. It was quite insightful to learn from the chair that organizing the conference is a tightrope walk with a good fraction of ticket sales occurring towards the last week and booking the venue has numerous constraints! Hats off!
As for my talk, I believe I did my part. Judging from my interaction, it
seemed that a lot of folks were oblivious to fact that
numpy is not so fast
for CPU-bound problems. I received some interesting questions as well:
Add some nice fluid dynamics visualizations?
There is a video on my Software page.
Can you use std. library
typingfor type annotations? Say for lists and dictionaries etc. This would allow for compatibility with
mypyand later on
mypycwhen it is ready.
This is seriously followed up as a possible
Can you accelerate Pandas?
It seems to be possible, especially when you follow functional programming style.
Can you accelerate OpenCV code?
This is a hard one. Of course, if you are dealing with images which are
numpy arrays, it is possible. I haven't seen any Python extensions
written (in Cython or other) to accelerate OpenCV code and even if it does
transonic is not designed to interface with libraries.
Can Pythran replace Cython / f2py for interfacing with native code?
Pythran does sound like Fortran, but it has nothing much in common - except the fact that Pythran and Fortran target scientific computing and HPC. Generally speaking, Pythran is meant to extend not interface.
However, Pythran can do some useful things such as generate C++ only code, free
from any runtime Python dependencies (using
-e argument) or even export
which is compatible with
The journey continues ...
As developers of project
transonic, we are hoping that you would try out the
project and adopt it in your personal scripts, notebooks and possibly in
packages that you develop. If you found
transonic useful, help us, encourage
us, by starring the project on GitHub
and sharing your