Shiny for Python
I am sure many data scientists out there are super-amped that Shiny for Python has moved from alpha to general availability (yep 
I am definitely one of them!).
Sure R has its plus points - I am looking at you ggplot2, my first data viz-package love 😉 - but as a major Python fangirl, making Shiny applications was not as fun as it should have been. Enter stage left Shiny for Python, for a potentially full back-end to front-end Python experience.
Taking Shiny for Python for a spin
Fortunately, I didn’t have to wait long for a reason to give it a go. I had the perfect use case: a simple dashboard to visualise metrics stored in a database.
Requirements:
☑ the app had to be hosted for free (very important)
☑ it had to be hosted somewhere that complies with organisational policies.
☑ it had to be quick and simple to build
☑ It could not be in R (OK, fine, that one was just my requirement).
Note: Streamlit is another extremely simple-to-write Python front-end framework, but Joe Cheng talks more about some reasons as to why not to use Streamlit (amongst other things) in an excellent talk here.
First thoughts
With an organisational subscription to shinyapps.io, using Python-flavour Shiny made sense. I got cracking on the app development using the excellent resources at Shiny for Python (or PyShiny), and pretty quickly had the bare bones of the app. My initial observations of the process were:
- developing in VSCode (using the Shiny for Python extension) was really straightforward - a ‘simple browser’ opens up in VSCode when you run your app and automatically refreshes when you make any changes. Because VSCode supports multiple languages, this is also a good IDE for CSS and JavaScript (more on that shortly).
- the documentation was good, but the fact that this is still a relatively immature product meant that some content was out of date.
- similarly, if you are expecting the bells and whistles that come with RShiny, you might be a bit disappointed. Currently, there is no
shinycssloadersorShinyWidgets… basically, you need to be super comfortable with CSS and JavaScript to really customise your dashboard. - in my opinion the code is more austere and stripped back than RShiny in a really good way: on the ui side, there is a series of nested function calls which mirror the shape of the dashboard. Server-side, it is simple to implement reactivity (using decorators).
Deployment madness 🚀
OK so a few drawbacks, but in general, so far, so good. I can deploy this app locally, it looks fantastic, everything works fine, happy days.
Except when it comes to publishing my app on shinyapps.io. Then things get a bit messy and some of the issues I mentioned above (immature product, some gaps in documentation) really come to the fore.
Challenge 1: .env file shenanigans
The instructions shinyapps.io supplies provides a step-by-step guide for setting up rsconnect. Except it omits quite an important detail - the .env file MUST be named - think of it as an anti-Voldemort. ALL of the thanks to my colleague for pointing this out to me and sparing me that pain of figuring this out myself, which might have happened never 😐.
Challenge 2: database drivers
When using RShiny and RStudio, connecting to a database (in this case an MS SQL Server database) in the app locally and on shinyapps.io is straightforward; using odbc and DBI packages means that you can use Posit’s drivers. However, it turns out sqlalchemy and / or pyodbc will not be able to access these drivers. A pretty big problem when all the data I need is in this database.
I thought the gig was up… thanks Shiny for Python, it was nice while it lasted. However, my stackoverflow-doomscrolling eventually bore fruit and the solution presented itself. No driver installation required! I could use a package I hadn’t come across before: pymssql, which builds on top of FreeTDS.
Fin 😰
One fully deployed app later, and my final thoughts are thus: Shiny for Python is fantastic for creating quick-and-dirty apps (probably for internal users), but it has a bit of a way to go to scale the heights of RShiny. Hopefully the userbase will take off and all the frills that RShiny currently enjoys will come to the Python-side.