Visualizers
See how machine learning works
Interactive explainers for the algorithms behind machine learning. Place the data yourself, change the parameters, and watch what the algorithm does with them \u2014 including what it does when you give it bad settings, which is usually where the understanding happens.
Built to be worked through, not watched. Each one is a real implementation of the algorithm it describes, running on data you provide, with the parameters exposed — including the ones that break it. Every page also carries a written explanation of what the algorithm is and where the demo simplifies things, because a diagram on its own teaches nobody anything.
4 visualizers
Optimisation
How models actually learn
Every paradigm below is trained by the same underlying process: follow the gradient downhill and hope the surface cooperates. This is that surface, in three dimensions, with the optimisers walking it.
Supervised learning
Supervised learning
Learning from labelled examples — the data arrives with the right answers attached, and the model's job is to find a rule that reproduces them on inputs it has not seen.
Unsupervised learning
Unsupervised learning
Learning with no labels at all. There is no right answer to reproduce, so the algorithm looks for structure that is already in the data — groups, densities, directions of variation.
Reinforcement learning
Reinforcement learning
Learning from consequences. No labelled examples and no dataset — an agent acts, receives a reward or nothing at all, and has to work out which of its earlier choices deserve the credit.