Overview

Over the past few months I’ve been working through a series of challenges (kata) put together by the folks at Agile Geosciences. These are small, geoscience-oriented puzzles that are great for practicing basic Python skills and keep you thinking. I’ve also been doing a lot of my coding lately in Streamlit as opposed to Jupyter Notebooks (or Lab). This has been a good learning experience as Streamlit requires a bit more structured thinking in how you put your code together compared to Jupyter, where you can jump from cell to cell. Below I’ll give a quick overview of my impressions of Streamlit and also give links to my Agile Geocomputing kata example solutions.

If you want to jump right into examples, you can find go look at my repository. If you copy the raw Github link for a given .py file, then you can actually run the Streamlit code directly from the web. Try running the example below with an active environment that has Streamlit installed:

streamlit run https://raw.githubusercontent.com/armstrys/Kata-Geochallenge-Solutions/master/streamlit-dashboards/04-Kata-Prospecting.py

These were fun to put together and though the puzzles not inherently interactive, I tried to add a few buttons, sliders, and dropdowns along the way just to keep things interesting.

The Streamlit Gallery is another good resource to see how community users are implementing projects in Streamlit. These examples show far more in-depth use-cases and the power of interactivity that Streamlit enables.

An Overview of Streamlit

Streamlit is a really interesting product. It is a totally different take from Jupyter Notebooks on how to interact with your code. That has some big advantages, but a few drawbacks as well.

Pros

  • Incredibly intuitive widgets. They just work. One line of code to assign a widget to a variable that can be used like any other variable.
  • Code runs top to bottom every time it’s executed, which is on save or by widget changes. This means no forgetting to re-run a cell in your notebook or accidentally re-running something that takes forever.
  • Time-intensive code can be cached by placing @st.cache() above a function. This is a key optimization, because the way Streamlit executes your code.
  • Can be used with any text editor.
  • Designed to create applets quickly and easily - again, this is different than Jupyter Notebooks by design and it makes building clean interfaces with interactive widgets very straight-forward.

Cons

  • More difficult to quickly test code and see results. The nature of the interface is not as natural as a Jupyter Notebook where each line of code spits out the result on the line below.
  • Cannot render onto platforms like Github the same way you could a .ipynb.
  • Finally, a bit more of a learning curve. This might not be as much of an issue for more experience coders. I found the friendly and forgiving sandbox of Jupyter Notebooks fundamentally easier to understand when getting started.

Initially, I planned to list a lack of third-party extensions on the Streamlit cons, but they have recently added a new feature called Components that allows for the addition of new components via custom HTML or Javascript. I don’t have a good handle on what the limits or possibilities of this addition, but you can already view some of the early contributions using the new feature in their components gallery. If you haven’t tried out Streamlit yet I hope this was enough useful info to convince you that you should give it a go (or not)!

Conclusion

These puzzles and learning different coding styles in Streamlit have been a great help to my personal Python development over the past few months. While I will probably stick mostly with Jupyter Notebooks and Jupyter Lab due to the flexibility and interoperability with multiple platforms, Streamlit will be my go-to tool any time I have a simple POC that is better served as an applet than a live document.