KindXiaoming/pykan ? reverse-engineered prompt

Build me a Python package for Kolmogorov Arnold Networks, with a simple notebook first and then a few example tutorials that show how to train them on small problems like function fitting and basic science style tasks.

I want the model to be easy to use from a notebook, with a quickstart example that trains a KAN, plots the learned curves, and shows how it compares to a normal neural net. Include the main features from the README, like training on CPU, a speed mode for when I do not need symbolic math, pruning for simpler models, and a way to try symbolic regression or interpret the network. Please make the code feel practical for experimentation, with clear plots and easy to tweak settings like width, grid size, depth, and regularization.

If you need to, look up the current docs and examples online so the behavior matches the project’s intended usage.