Hi everyone,
I love the playground, but I noticed it's missing a great opportunity to teach students how classification boundaries are actually just approximated mathematical functions.
I’d love to suggest adding a "Mathematical Functions" dataset to the classification panel. The approach fits perfectly into the existing playground code and UI:
- Blue dots: Any random points generated above a given function line (y > f(x)).
- Orange dots: Any random points generated below the function line (y < f(x)).
By using raw X and Y coordinates, students could watch the hidden layers take straight lines and miniature linear splits and literally bend them to mimic a curve.
I built a quick proof-of-concept for this in a Google Colab notebook using a Sine Wave as the boundary. You can see how cleanly a simple network approximates the wave here:
https://colab.research.google.com/drive/1lC0-8LdycWm59QfXsh-pySWyFz2mnwoV?usp=sharing
While my notebook shows a sine wave, this dataset idea could easily be generalized to a bunch of other classic functions, like a Cubic function (y = x^3 - x) or a V-shape (y = |x|) to show how networks handle sharp bends.
I think this would be a massive educational upgrade for bridging algebra with neural networks. Let me know what you think!
Hi everyone,
I love the playground, but I noticed it's missing a great opportunity to teach students how classification boundaries are actually just approximated mathematical functions.
I’d love to suggest adding a "Mathematical Functions" dataset to the classification panel. The approach fits perfectly into the existing playground code and UI:
By using raw X and Y coordinates, students could watch the hidden layers take straight lines and miniature linear splits and literally bend them to mimic a curve.
I built a quick proof-of-concept for this in a Google Colab notebook using a Sine Wave as the boundary. You can see how cleanly a simple network approximates the wave here:
https://colab.research.google.com/drive/1lC0-8LdycWm59QfXsh-pySWyFz2mnwoV?usp=sharing
While my notebook shows a sine wave, this dataset idea could easily be generalized to a bunch of other classic functions, like a Cubic function (y = x^3 - x) or a V-shape (y = |x|) to show how networks handle sharp bends.
I think this would be a massive educational upgrade for bridging algebra with neural networks. Let me know what you think!