I recently made a set of explorable animations that takes an in-depth look at the humble perceptron. Aside from the typically shown linearly separable case, you can also explore convergence with an adjustable margin, as well as the Maxover algorithm and Voted Perceptron variants, which offer good performance, even when the data isn't linearly separable. There was very little information online about the Maxover algorithm, save for the one paragraph on the Wikipedia page. And the paper itself was pretty annoying to read through. I'm pretty glad that I was able to get the messy pseudocode from the paper into something visual you can run in your browser.