Decision Boundary Shiny App

Hey there! I made this Shiny for Python app over the summer that shows the decision boundaries for SUPER simple ML models so my students can see how different models create different types of decision boundary shapes.

I currently have 3 data sets (Iris, Palmer Penguins, and Diabetes), but I’d love some suggestions for data sets with differently shaped decision boundaries (REAL DATA ONLY). Any suggestions?

(See the “Shiny for Python Decision Boundary App” link here: Coding because the links to Shiny for Python apps are SUPER long. )

1 Like

I don’t have any data suggestions (yet!) but holy heck those graphs are really cool! I really like how you can see how each model comes up with slightly different boundaries.

are there any things to look for in data sets that would kind of force models to come up with drastically different boundaries?

1 Like

That’s such a good question! Basically any data that’s not linearly separable. If you look at the plots of the current data, even though there’s some overlap between groups, you could roughly draw a few straight lines and get a pretty good decision boundary.

But things with curved, irregularly shaped, or just weirdly shaped groups (e.g. something that’s like the top two rows here:
image
)

Those are extreme examples, but that sort of thing!:partying_face:

1 Like